Query by Humming, Melody Representation, Hidden Markov Model (26.05.11)
Formal Metadata
Title 
Query by Humming, Melody Representation, Hidden Markov Model (26.05.11)

Title of Series  
Part Number 
9

Number of Parts 
14

Author 

Contributors 

License 
CC Attribution  NonCommercial 3.0 Germany:
You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal and noncommercial purpose as long as the work is attributed to the author in the manner specified by the author or licensor. 
Identifiers 

Publisher 

Release Date 
2011

Language 
English

Producer 

Production Year 
2011

Production Place 
Braunschweig

Content Metadata
Subject Area  
Abstract 
In this course, we examine the aspects regarding building multimedia database systems and give an insight into the used techniques. The course deals with contentspecific retrieval of multimedia data. Basic issue is the efficient storage and subsequent retrieval of multimedia documents. The general structure of the course is:  Basic characteristics of multimedia databases  Evaluation of retrieval effectiveness, PrecisionRecall Analysis  Semantic content of imagecontent search  Image representation, lowlevel and highlevel features  Texture features, randomfield models  Audio formats, sampling, metadata  Thematic search within music tracks  Query formulation in music databases  Media representation for video  Frame / Shot Detection, Event Detection  Video segmentation and video summarization  Video Indexing, MPEG7  Extraction of lowand highlevel features Integration of features and efficient similarity comparison  Indexing over inverted file index, indexing Gemini, R * trees

00:00
Algorithm
Multiplication sign
3 (number)
Database
Mathematical model
Neuroinformatik
Information retrieval
Query language
Representation (politics)
Multimedia
Aerodynamics
Metropolitan area network
Area
Algorithm
Pattern recognition
Matching (graph theory)
Video tracking
Database
Line (geometry)
Mathematical model
Markov chain
Computer animation
Information retrieval
Network topology
Knowledge representation and reasoning
Multimedia
Energy level
Musical ensemble
02:00
Point (geometry)
Complex (psychology)
Trail
Line (geometry)
Multiplication sign
Image resolution
Control flow
Computer file
Insertion loss
Bit rate
2 (number)
Information retrieval
Frequency
Order (biology)
Goodness of fit
Bit rate
Labour Party (Malta)
Band matrix
Vector graphics
Query language
Information
Metropolitan area network
God
Physical system
Structural load
Complex (psychology)
Counting
Database
Bit
Line (geometry)
Statistics
Measurement
Band matrix
Sample (statistics)
Computer animation
Oval
Right angle
Musical ensemble
Whiteboard
06:28
Slide rule
Length
Multiplication sign
Execution unit
Sheaf (mathematics)
Mereology
Number
Information retrieval
Frequency
Mathematics
Different (Kate Ryan album)
Semiconductor memory
Query language
Process (computing)
Bounded variation
Hydraulic jump
Metropolitan area network
Scale (map)
Scaling (geometry)
Type theory
Computer animation
Information retrieval
Different (Kate Ryan album)
Interpreter (computing)
Musical ensemble
Bounded variation
10:47
Decision theory
Multiplication sign
Database
Mathematical analysis
Mereology
Mathematical model
Rule of inference
Machine code
Variance
Information retrieval
Goodness of fit
Envelope (mathematics)
Term (mathematics)
Query language
Codierung <Programmierung>
Pairwise comparison
Amenable group
Beta function
Differential (mechanical device)
Computer file
Video tracking
Database
Mathematical model
Computer animation
Customer relationship management
Resultant
Row (database)
13:07
Cue sports
Point (geometry)
Computer file
State of matter
Multiplication sign
Calculation
Electronic mailing list
Limit (category theory)
Plastikkarte
Distance
Fast Fourier transform
Machine code
Hypothesis
Neuroinformatik
Architecture
Wave
Mathematics
Goodness of fit
Flow separation
Knowledge representation and reasoning
Different (Kate Ryan album)
Representation (politics)
Multimedia
output
Metropolitan area network
Scalable Coherent Interface
Key (cryptography)
Server (computing)
Electronic program guide
Client (computing)
Database
Machine code
Flow separation
Distance
Computer animation
Thermal radiation
Phase transition
Compilation album
Musical ensemble
Electronic visual display
Resultant
Row (database)
Cloning
Tunis
17:38
Point (geometry)
Frame problem
Group action
Existence
Computer file
Network operating system
Multiplication sign
Disk readandwrite head
Mereology
Wave packet
2 (number)
Word
Mathematics
Virtual reality
Different (Kate Ryan album)
Noise
Data structure
output
Metropolitan area network
Scripting language
Pattern recognition
Scaling (geometry)
Matching (graph theory)
Key (cryptography)
Sampling (statistics)
Virtualization
Bit
Digital signal
Frame problem
Sample (statistics)
Computer animation
Mathematical singularity
Data conversion
Right angle
Musical ensemble
Quicksort
Data structure
Row (database)
22:06
Multiplication sign
Insertion loss
Mereology
Mathematics
Exception handling
Algorithm
Structural load
Sampling (statistics)
Sound effect
Maxima and minima
Bit
Fast Fourier transform
Type theory
Sample (statistics)
Frequency
Uniformer Raum
Different (Kate Ryan album)
Data conversion
Volume
Web page
Frame problem
Spectrum (functional analysis)
Calculation
Maxima and minima
Control flow
Average
Graph coloring
Product (business)
Number
Frequency
Finite element method
Causality
Term (mathematics)
Average
Boundary value problem
Hydraulic jump
Metropolitan area network
Physical law
Neighbourhood (graph theory)
Length
Volume (thermodynamics)
Frame problem
Carry (arithmetic)
Singleprecision floatingpoint format
Number
Computer animation
Function (mathematics)
Radiofrequency identification
Musical ensemble
Boundary value problem
Hydraulic jump
Spectrum (functional analysis)
30:42
Point (geometry)
Computer virus
Pixel
Line (geometry)
Image resolution
Decision theory
MIDI
Set (mathematics)
Online help
Disk readandwrite head
Machine code
Sequence
Chaining
Knowledge representation and reasoning
Different (Kate Ryan album)
Chain code
Position operator
Metropolitan area network
Physical system
Pattern recognition
Sampling (statistics)
Planning
Machine code
Line (geometry)
System call
Chaining
Computer animation
Knowledge representation and reasoning
System programming
Representation (politics)
Reading (process)
35:37
Scale (map)
Scaling (geometry)
Characteristic polynomial
Multiplication sign
Line (geometry)
Ordinary differential equation
Mereology
Machine code
Chaining
Mathematics
Goodness of fit
Computer animation
Octave
Query language
Energy level
Representation (politics)
output
Row (database)
38:21
Classical physics
Differential (mechanical device)
Maxima and minima
Directory service
Infinity
Special unitary group
Machine code
Arm
Field (computer science)
Valueadded network
Insertion loss
Musical ensemble
Physical law
Row (database)
Gamma function
Summierbarkeit
Algebra
Tunis
Metropolitan area network
Execution unit
Raw image format
Uniqueness quantification
Bit
Machine code
Subject indexing
LangevinGleichung
Malware
Computer animation
Personal digital assistant
Uniform resource name
System identification
Musical ensemble
Tunis
Wide area network
40:43
Email
Wechselseitige Information
User interface
State diagram
Demo (music)
Programmable readonly memory
WebDAV
Graphic design
Special unitary group
Area
Pointer (computer programming)
Uniformer Raum
Insertion loss
Forest
Electronic meeting system
Physical law
Row (database)
LipschitzStetigkeit
Metropolitan area network
Unitäre Gruppe
Link (knot theory)
View (database)
Interior (topology)
IBM Systems Application Architecture
Amsterdam Ordnance Datum
Mass
Mereology
Trigonometric functions
Annulus (mathematics)
Uniform resource name
Computer music
Personal area network
Simulation
Optical disc drive
Router (computing)
Wide area network
Polygon mesh
Personal identification number
Maxima and minima
3 (number)
Division (mathematics)
Infinity
Plot (narrative)
Valueadded network
Wave
Arithmetic mean
Computer worm
Quantum
Conditionalaccess module
Maß <Mathematik>
World Wide Web Consortium
Newton's law of universal gravitation
Information management
Suite (music)
Greatest common divisor
Lemma (mathematics)
Coma Berenices
Linear programming
Core dump
Binary file
Frame problem
String (computer science)
Large eddy simulation
Normed vector space
Boom (sailing)
Bus (computing)
Musical ensemble
Game theory
Capability Maturity Model
Differential (mechanical device)
Structural equation modeling
Arm
10 (number)
Nominal number
Information retrieval
Duality (mathematics)
Mathematics
Graphical user interface
Moving average
Drum memory
Addressing mode
Chisquared distribution
Menu (computing)
Multiagent system
Computer cluster
Hill differential equation
Convex hull
Damping
Impulse response
Software engineering
Asynchronous Transfer Mode
MUD
Service (economics)
Line (geometry)
Artificial neural network
MIDI
Directory service
Dynamic randomaccess memory
Discrete element method
Law of large numbers
Hand fan
Emulation
Linear multistep method
Musical ensemble
Interrupt <Informatik>
Maize
Gamma function
Arithmetic logic unit
Summierbarkeit
Atomic nucleus
Window
Execution unit
Raw image format
Electronic data interchange
Sine
Magnetooptical drive
Computer data logging
Order of magnitude
Color management
Local area network
Plane (geometry)
CAN bus
Inclusion map
Subject indexing
Number
Computer animation
Intrusion detection system
Lie group
Royal Navy
Form (programming)
Identity management
Computerassisted translation
42:53
User interface
Likelihoodratio test
Special unitary group
Predictability
Exploratory data analysis
Pointer (computer programming)
Uniformer Raum
Insertion loss
Quilt
Row (database)
Physical law
LipschitzStetigkeit
God
Metropolitan area network
Link (knot theory)
View (database)
Interior (topology)
Amsterdam Ordnance Datum
Hidden Markov model
Pi
Uniform resource name
Theorem
Personal area network
Computer music
Wide area network
Ewe language
Personal identification number
Bending
Maxima and minima
3 (number)
Division (mathematics)
Infinity
Computerintegrated manufacturing
Valueadded network
Conditionalaccess module
Newton's law of universal gravitation
Information management
Coma Berenices
Usability
Grand Unified Theory
Binary file
Latent class model
Mathematics
Musical ensemble
Capability Maturity Model
Manufacturing execution system
Structural load
Icosahedron
Counting
Arm
Thermische Zustandsgleichung
Variance
Duality (mathematics)
Moving average
Addressing mode
Library (computing)
Arc (geometry)
Repetition
Network switching subsystem
Menu (computing)
4 (number)
Convex hull
Hill differential equation
Software engineering
Rounding
MUD
Line (geometry)
MIDI
Artificial neural network
Directory service
Limit (category theory)
Electronic mailing list
Host Identity Protocol
Hand fan
Law of large numbers
Emulation
Local Group
Writing
Maximum likelihood
Summierbarkeit
Loop (music)
Execution unit
Raw image format
Torus
Sine
Online help
Magnetooptical drive
Computer data logging
CAN bus
Inclusion map
Number
Coding theory
Computer animation
Lie group
5 (number)
Identity management
Cloning
44:55
Multiplication sign
Execution unit
Database
Counting
Likelihoodratio test
Mereology
Arm
Machine code
Pointer (computer programming)
Mathematics
Hash function
Query language
Row (database)
Physical law
Drum memory
Position operator
Physical system
Metropolitan area network
Web page
Network switching subsystem
Point (geometry)
Menu (computing)
Distance
Uniform resource name
Mathematical singularity
Personal area network
Hill differential equation
Convex hull
Quicksort
Reading (process)
Wide area network
MUD
Artificial neural network
Maxima and minima
Directory service
Limit (category theory)
Infinity
Emulation
Valueadded network
Linear multistep method
Local Group
Writing
Frequency
Goodness of fit
String (computer science)
Zeitinvariantes System
Musical ensemble
Summierbarkeit
Newton's law of universal gravitation
Domain name
Execution unit
Raw image format
Information management
Matching (graph theory)
Database
Grand Unified Theory
Line (geometry)
Subject indexing
Number
Computer animation
Information retrieval
Network topology
Labour Party (Malta)
Musical ensemble
Tunis
50:12
Transformation (genetics)
Multiplication sign
Maxima and minima
Open set
Mereology
Distance
Mathematical model
Machine code
Number
Sequence
Measurement
Term (mathematics)
Operator (mathematics)
String (computer science)
Pairwise comparison
Operations research
Matching (graph theory)
Scaling (geometry)
Maxima and minima
Bit
Database
Measurement
Distance
Similarity (geometry)
Error message
Computer animation
String (computer science)
output
Musical ensemble
Escape character
Resultant
54:39
Point (geometry)
Table (information)
Convex hull
Transformation (genetics)
Graph (mathematics)
Multiplication sign
Online help
Insertion loss
Heat transfer
Xray computed tomography
Mereology
Distance
Machine code
Sequence
Web 2.0
Mathematics
Degree (graph theory)
Strategy game
Causality
Different (Kate Ryan album)
Military operation
String (computer science)
Operator (mathematics)
Green's function
Computerassisted translation
Operations research
Matching (graph theory)
Polygon mesh
Block (periodic table)
Repetition
Database
Maxima and minima
System call
Singleprecision floatingpoint format
Word
Computer animation
Finite difference
String (computer science)
Quicksort
Musical ensemble
Table (information)
Row (database)
1:04:20
Dynamical system
Transformation (genetics)
Graph (mathematics)
Transport Layer Security
Multiplication sign
Mereology
Distance
Machine code
Computer programming
Neuroinformatik
Programmer (hardware)
Sign (mathematics)
Mathematics
Operator (mathematics)
Vertex (graph theory)
Aerodynamics
Data structure
Operations research
Algorithm
Link (knot theory)
Graph (mathematics)
Graph (mathematics)
Bit
Computer programming
Number
Computer animation
Personal digital assistant
1:07:39
Point (geometry)
Group action
State of matter
Multiplication sign
Complementarity
Ultraviolet photoelectron spectroscopy
Similarity (geometry)
Equivalence relation
Machine code
Frequency
Mathematics
Insertion loss
Different (Kate Ryan album)
Operator (mathematics)
output
Error message
Condition number
Operations research
Pairwise comparison
Matching (graph theory)
Structural load
Closed set
Line (geometry)
Equivalence relation
Error message
Computer animation
output
1:10:51
Point (geometry)
Dynamical system
Group action
Transformation (genetics)
Latin square
Graph (mathematics)
Multiplication sign
Direction (geometry)
Transport Layer Security
Electronic program guide
Control flow
Set (mathematics)
Streaming media
Mereology
Machine code
Computer programming
Sequence
Programmer (hardware)
Sign (mathematics)
Different (Kate Ryan album)
String (computer science)
Query language
Vertex (graph theory)
Physical system
Window
Operations research
Greedy algorithm
Time zone
Graph (mathematics)
Matching (graph theory)
Relational database
Graph (mathematics)
Interior (topology)
Independence (probability theory)
Computer programming
Sequence
Computer animation
Video game
Right angle
Serviceoriented architecture
Metric system
Family
Resultant
Window
1:21:13
Point (geometry)
Group action
Finitismus
Codierung <Programmierung>
Multiplication sign
MIDI
Database
Machine code
Order of magnitude
Number
Frequency
Mathematics
Different (Kate Ryan album)
Chain code
Linear programming
Query language
Information
Scale (map)
Shift operator
Cellular automaton
Point (geometry)
Sound effect
Bit
Machine code
Mathematics
Number
Computer animation
Thermal conductivity
Window
1:23:33
Distribution (mathematics)
Transformation (genetics)
Codierung <Programmierung>
Multiplication sign
Database
Ordinary differential equation
Mereology
Number
Mathematics
Goodness of fit
Different (Kate Ryan album)
Symmetric matrix
Hydraulic jump
Distribution (mathematics)
Matching (graph theory)
Cellular automaton
Database
Machine code
Measurement
Message passing
Process (computing)
Computer animation
Mathematical singularity
Musical ensemble
Hydraulic jump
1:28:30
Frame problem
Focus (optics)
Matching (graph theory)
Codierung <Programmierung>
1 (number)
Sound effect
Control flow
Database
Insertion loss
Matching (graph theory)
Distance
Proper map
Machine code
Frame problem
Distance
Computer animation
Knowledge representation and reasoning
Query language
Active contour model
Representation (politics)
Musical ensemble
Hydraulic jump
Hydraulic jump
Physical system
1:30:35
Point (geometry)
Frame problem
Multiplication sign
Time series
Database
Mereology
Machine code
Knowledge representation and reasoning
Different (Kate Ryan album)
Query language
Active contour model
Representation (politics)
Metropolitan area network
Area
Scripting language
Pairwise comparison
Matching (graph theory)
Structural load
Java applet
Database
Frame problem
Discounts and allowances
Type theory
Event horizon
Computer animation
Network topology
Knowledge representation and reasoning
Representation (politics)
1:33:37
Dynamical system
Multiplication sign
Database
Mereology
Machine code
Measurement
Information retrieval
Mathematics
Object (grammar)
Military operation
Query language
Active contour model
Information
Aerodynamics
Pairwise comparison
Closed set
Point (geometry)
Moment (mathematics)
Image warping
Sampling (statistics)
Bit
Maxima and minima
Sequence
Distance
Data mining
Series (mathematics)
Knowledge representation and reasoning
Quicksort
Representation (politics)
Resultant
Point (geometry)
Frame problem
Time series
Student's ttest
Distance
Sequence
Frequency
Clifford algebra
Alphabet (computer science)
Data mining
Representation (politics)
Pairwise comparison
Matching (graph theory)
Length
Frame problem
Number
Computer animation
Alphabet (computer science)
Information retrieval
Family
1:36:46
Point (geometry)
Axiom of choice
Dynamical system
Transformation (genetics)
State of matter
Multiplication sign
1 (number)
Time series
Matching (graph theory)
Distance
Traverse (surveying)
Computer programming
Linear multistep method
Information retrieval
Pointer (computer programming)
CAN bus
Different (Kate Ryan album)
Green's function
Circle
Aerodynamics
Scripting language
Covering space
Pairwise comparison
Curve
Matching (graph theory)
Image warping
Plastikkarte
Maxima and minima
Total S.A.
Machine code
Line (geometry)
Complete metric space
Measurement
Computer animation
Personal digital assistant
Series (mathematics)
Order (biology)
Video game
Quicksort
Game theory
Whiteboard
Local ring
Row (database)
1:48:56
Point (geometry)
Dynamical system
Multiplication sign
Decision theory
Calculation
Matching (graph theory)
Distance
Computer programming
Information retrieval
Different (Kate Ryan album)
Personal digital assistant
KerrLösung
Boundary value problem
Analytic continuation
Curve
Key (cryptography)
Relational database
Network switching subsystem
Total S.A.
Lattice (order)
Computer programming
Computer animation
Personal digital assistant
Right angle
Game theory
Boundary value problem
1:53:18
Metropolitan area network
Dynamical system
Satellite
Multiplication sign
Lemma (mathematics)
Maxima and minima
Menu (computing)
Bit
Port scanner
Computer programming
Emulation
Computer animation
Intrusion detection system
Semiconductor memory
Series (mathematics)
Inverter (logic gate)
Speech synthesis
Physical law
Musical ensemble
Quicksort
Summierbarkeit
Wide area network
1:55:22
Email
Diagonal
Multiplication sign
Lattice (order)
Sequence
Computer programming
Demoscene
Frequency
Computer animation
Oval
Information systems
Game theory
Whiteboard
Window
1:57:42
Point (geometry)
Dynamical system
Length
Scaling (geometry)
Multiplication sign
Calculation
Maxima and minima
Time series
Matching (graph theory)
Distance
Mereology
Disk readandwrite head
Variable (mathematics)
Cartesian coordinate system
Area
Different (Kate Ryan album)
Aerodynamics
Summierbarkeit
Local ring
Position operator
Area
Time zone
Scaling (geometry)
Physical law
Image warping
Length
Uniform convergence
Variable (mathematics)
Degree (graph theory)
Computer animation
Uniformer Raum
Series (mathematics)
Schmelze <Betrieb>
Speech synthesis
Diagonal
Matrix (mathematics)
Local ring
Wide area network
2:02:11
Point (geometry)
Dynamical system
Multiplication sign
Price index
Database
Mereology
Frequency
Dedekind cut
Different (Kate Ryan album)
Personal digital assistant
Subject indexing
Query language
Aerodynamics
Local ring
Physical system
Area
Matching (graph theory)
Cliquewidth
Kolmogorov complexity
Image warping
Bit
Measurement
Distance
Computer animation
Series (mathematics)
Envelope (mathematics)
Object (grammar)
Musical ensemble
Data structure
Freezing
Extension (kinesiology)
2:04:38
Point (geometry)
Frame problem
Randomization
Transformation (genetics)
State of matter
Multiplication sign
Time series
Set (mathematics)
Shape (magazine)
Mereology
Mathematical model
Event horizon
Mathematical model
Reduction of order
Normal (geometry)
Metropolitan area network
Physical system
Electric generator
Structural load
Physical law
Maxima and minima
Basis <Mathematik>
Line (geometry)
Transformation (genetics)
Frame problem
Computer file
Markov chain
Mathematical model
Shooting method
Process (computing)
Event horizon
Computer animation
Personal digital assistant
Series (mathematics)
Information retrieval
Musical ensemble
Form (programming)
Resultant
2:09:02
Discrete group
Point (geometry)
Implementation
Group action
State of matter
Multiplication sign
Time series
Infinity
Mereology
Mathematical model
Event horizon
Independence (probability theory)
Sequence
Chaining
Atomic number
String (computer science)
Set (mathematics)
Energy level
Implementation
Physical system
Electric generator
State of matter
Total S.A.
Mathematical model
Markov chain
Chaining
Event horizon
Stochastic
Computer animation
Alphabet (computer science)
Personal digital assistant
Phase transition
Envelope (mathematics)
Energy level
Resultant
Reading (process)
2:12:39
Point (geometry)
Distribution (mathematics)
Distribution (mathematics)
State of matter
Weight
Multiplication sign
Prisoner's dilemma
Characteristic polynomial
State of matter
Markov chain
Set (mathematics)
Markov chain
Singleprecision floatingpoint format
Invariant (mathematics)
Computer animation
Personal digital assistant
Zeitinvariantes System
Process (computing)
Condition number
2:15:59
Point (geometry)
Web page
State of matter
View (database)
Multiplication sign
Mereology
Mathematical model
2 (number)
Sequence
Chaining
Envelope (mathematics)
Different (Kate Ryan album)
Energy level
Structural load
Physical law
State of matter
Sequence
Shape (magazine)
Frame problem
Singleprecision floatingpoint format
Event horizon
Computer animation
Personal digital assistant
Website
Musical ensemble
Routing
Family
2:20:19
Point (geometry)
State observer
Group action
Graphics tablet
Observational study
Computer file
State of matter
Infinity
Mereology
Event horizon
2 (number)
Mathematics
Goodness of fit
Envelope (mathematics)
Operator (mathematics)
Conditionalaccess module
State of matter
Sequence
Frame problem
Mathematical model
Markov chain
Computer animation
Personal digital assistant
Video game
Social class
Summierbarkeit
Bounded variation
2:25:59
Point (geometry)
State observer
State of matter
Multiplication sign
Programmable readonly memory
Markov chain
Mathematical model
Event horizon
2 (number)
Sequence
Causality
Operator (mathematics)
Authorization
Uniform boundedness principle
Random variable
Condition number
State of matter
Special unitary group
Hidden Markov model
Frame problem
Mathematical model
Event horizon
Computer animation
Order (biology)
Pattern language
Probability density function
Conditional probability
2:29:53
Point (geometry)
Neighbourhood (graph theory)
Random number
Group action
Distribution (mathematics)
State of matter
Texture mapping
Multiplication sign
Markov chain
Variable (mathematics)
Sequence
Pi
Process (computing)
Category of being
Random variable
Distribution (mathematics)
State of matter
Sequence
Markov chain
Process (computing)
Stochastic
Computer animation
Personal digital assistant
Condition number
Musical ensemble
2:32:07
Probability distribution
Point (geometry)
State observer
Group action
Randomization
Statistics
Distribution (mathematics)
State of matter
Multiplication sign
Mathematical model
2 (number)
Web 2.0
Pointer (computer programming)
Invariant (mathematics)
Set (mathematics)
Process (computing)
Social class
Area
Distribution (mathematics)
State of matter
Exponential function
Line (geometry)
Hidden Markov model
Markov chain
Mathematical model
Process (computing)
Stochastic
Computer animation
Summierbarkeit
2:36:16
State observer
Convex hull
State of matter
Codierung <Programmierung>
Multiplication sign
MIDI
Maxima and minima
Markov chain
Solid geometry
Streaming media
Mereology
Mathematical model
Computer programming
Information retrieval
Different (Kate Ryan album)
Query language
Amenable group
Algorithm
Matching (graph theory)
Weight
Electronic mailing list
Streaming media
Bit
Total S.A.
Sequence
Frame problem
Markov chain
Mathematical model
Shooting method
Process (computing)
Computer animation
Order (biology)
Knowledge representation and reasoning
Finitestate machine
2:38:40
Lecture/Conference
00:00
it felt as allways on Thursday morning so much pleasure to welcome you to lecture about multimedia databases and then we finally moved to
00:12
the area of audio retrieval it to the world of sound and that last week you were singing brightly and focusing on the retrieving of recognising where the extra tried to expressed by you think and and it seemed to work what well timed this kind of you to the techniques of the of basically built around the man the of songs about the me lines of me pitches and there was basically were focusing on in the last lecture so were talking about tracking Algorithms for for pictures of how do you recognise what the OMX the prominent the most prominent note in a piece of music is at a certain time and pudding these notes together to for many line makes basically a mellowed the and this is how we and and the Tree for all recognise of audio pieces to date
01:24
we will go as the deeper into the audio retrieval we will again focus on Monday on pitch recognition fear about hummingbirds kind of 1 of the possibilities found will talk about the representation the full representation of melodies and matching of the so called the computer do that and very interesting and in the end we will walk to up all ballistic models will look at it mock most to kind of built melodies and and recognised sound of it
02:03
so I'm talking about statistical feets is about load of feats of like we did last week bomb resulted in and in the end global things global impressions of a piece of music for example allow the band with the brightness 0 within rate pitches and stuff like that but that is where a very nice and good but it's almost global skated so you can be the average brightness all brightness and pieces of of of musical 5 from that when we described the piece of new we basically do it by feature of actors and what of the problems about the future Vectra's is that they are static as opposed to music pieces which up nonstandard which up some variable in time and set up a new way basically seconds of feature Bektas 42 point in time and it's a matter of resolution many of those I have time for each point in time but it differently to recruit measuring these things still we do not get consider all the things that might be in for when you classified pieces of new the for example of the man that the lines so what was the guys singing the resemble the piece but you just take a point in time the measure of the value of the doesn't give you the feeling that the complex with the steady reasons many breaks in Italy which on the other hand makes a good distinction criteria for pieces of music by 2 when you a search for the music its also difficult because the only thing we came up with win focusing on the right is a kind of fear a by example because tried to sing something all we could try to to a together music he's lectures on those you know like recalled the music peace and take that as an example as future and and and then look for everything that a similar on the other hand you might well have period like I'm all my God that losses be from some upside in all like that at the time and you know now we don't know because a don't see the brightness the of all lightless all the bandwidth of something though everybody knows were means that makes it more interesting to focus on all it a bit more of fellow wasted discuss about that but the rhythm of what what makes the piece of music the piece of music of the visible what to do to database so
05:07
I'm we have to to due to discriminate between more complex pieces of music is if it is kind of some of the other information and mellow died is it is kind of what counts and a lack of everybody can go with the Miller and a whistle the to of some popular song about because the usually quite simple and from its board of often called you'll need no like it wants its way into your brain and you never get out of attic again and this is kind of kind of some the wait for New Labour the period despite by the high mingle with sling singing the melody and then the trail system should try to find the smell of the sea and now it comes somewhere in the music of the mind of the from the Stop the correct Modi it might be the rest for more of than not because of what this is what I notice of remember the celebrity speed spent the day in the lead the at at all my that it could look at it that way I'm well
06:32
so why everyone basically I have to do is we have to focus on what is now that the so basically it's kind of different it's just the change over time although to have a certain length of them change over time that's different kind of scale so we of the mind of major scale so we might have a lot of pieces of major from its may be slide variations of the human soul we very of the RUC number was made up of a very popular to of what history of prominent today is kind of about remixing so what it is to take its from the Eighties that everybody like a new put a beat unit and basically you contribution to the music of everybody will go like are because nobody remember we eighties and he more than when it was good in the Eighties that might be good no again and so more also slide variations of interpretations of of of the salt and might be of what might be might be developed from than the have different intervals of jumps of frequencies so the same kind of melody pump on a note of scale may occur in different ways so the soprano might stop singing the ability and some the Frankston not trust that in the end it was bright of the widow of the will go on sale in the jump in the middle of a ballot or you have a repetition of the melody in different voices which makes a difference to the difficult to find with about so how does the fear for me legs work out how the of new at the so you this is the prop up and if we can't do that if we can just the the melody off from the Modiano and then find a way to represent the mellowed the pieces in pieces of music such so that we can match it then we have become a good way to walks finding were we need of the
09:09
had a man of these not though that is in the eastern all the note from quencher both to like you have to be caucus for a palm and everybody's playing something not everybody is playing mellowed for example of the guys here with the bases just playing the and Blue of my fillet following over time but the hockey ever play the melody I'm with the Wilone's you may be sure they play part of the melody sometimes that some of piano whatever intellect can be kind of everything from sections usually never place amenity because of the on the 1st to play the memory and city of different signals in the co plaques piece of music and you need to detect which types of some of which notes basically make of the mentality for you take over time it's just that it's a times enough but for cost if we huge something we can sell it on the map of the for you want scandal that we kind of have a discriminate this this makes of different how to find out what the prominent men locked up in request of how we do that all week how Hopkin we do that and I'm gap so called we to take them out well
10:52
basically the crew is that we have to answer off what is a melody how can we represent Melodee's if we get them and how to beat the Kiwis policemen and we will usable based approach where because of model of the mellowed the and part of the mood the model at the end individual notes with the different envelopes to find out what as part of the by the and what is not but for all the mouth as good get so that all
11:35
said several of them Southwest out there in them in the muck of place and a new rule that the couple from time in the next the to the 1st we want to talk about how to actually recognise milady so you get the feeling off off what what what actually happened so the 1st step is for me late in the year and I couldn't can do that by any kind how you can transport amenities you my time you might whistle you might saying well you might take an audiophile say that the melee the so that some of the people see for you and then you have to extract than the interested in what to extract the melody from the recording and we can do a lot of things specialist the pitch tracking so the prominent note getting up a prominent about is that this kind of the interest in part from than you have to and code them and the different ways of doing the cost of frantic would come to that in a minute from end after the encoding you can competitive different coatings to the encodings that you already put into the database which may be of the difficult or the inefficient and so you have to make a decision somehow and then you return the results of what was most symbol in terms of the manager
13:09
so what basically happens is that you records something so somebody's thing about of and you recall it and get it into a computer representations which way file and what he can do that then do is 1 of you can do spectral and that this fast Fourier Transform you can get the pitches out of it if you can get the radiation's of the cellphone on to the pitch Prebilic of music hall many points and times to you have with the same pitch on for of like long notes to order that the change of tone and like chalk note of the new calculated some some some kind of of of representation cold we would use of both of them up and then you just put it into the database of care and the kind of like a compared the codes pulled out of phase with fewer coach that was extracted from using a from your from your humming Omni you can use any kind of distance where you can go wrong refused of code you will be stuck with the editing distant from also revisit that in a minute of then you say OK it which are the closest matches so what what was the best result of the new display at and very easy to get to the fund people from if you just saying it correct UK and this is the typical of phrase of the music of this kind of a man but the basic the whole you build Thesis the open but the end but
15:05
it is quite simple you can't saying you can time you can't was pulled up and depending on what you do is you get to the very different problems with because if people saying well some of them are singers some all them on that actually most of them are not so if some people try to use the and bad Cairo key but I'm even if they think they made hitting the right note is not so allways easy and so that the state the signal will have a strong individuality below might be mistakes in it for them might be I'm a sudden sloppiness and that's not really well done I'm having that was 1 of the 1st ideas about got big dating back to the early days of the of multimedia databases time because it is so it is kind of like a very the easy way to transport Limavady and everybody can do it in kind of the same waited on the special skilled 4 4 4 4 humming means that of the 1 of the more coming schools or something like that of 5 singing schools so should be there for a reason and I'm and a very off only to discriminate the notes limited that I'm you would he would take it sounds that can be ready for the computer for a time and when you go tucked up a Tom everybody knows what was caught it is even with me not being able to sing and a wooden no to sing a song not anyway of of of ideas with playing afraid very similar to humming many people can do it and do you have very little individuality because of the same kind of told that his status as a built in and in West playing and there you have very good notes separation because it at the with 6 stops and that takes on because you have to breeze at some point in of of the new breeze between notes enough and that makes for a very good separate well you could also
17:40
be so included by virtual instrument for example Greenstone ditch loving the and that of the year this small key boiled over here and then you could have and Blue along with the mouse under just pressed the keys and the for basic the melody and then you had to be called down here end of just stop the recalling press that he's and some although which made of notes and and would also make the duration from and could then basically a file it and put it into a new the correct so this is also a popular way of doing it without relying on the local skills all of behind skills of people
18:34
than what you do is you have to convert if you have to have a re caught the eye of the humming although with playing the singing and you get to do that with a quite low simply right because it if it it's not a book you are likely to find a point of the home on the palm my hand of after to fact that some people created saying this is not what I need for The matching this not the interesting part of the managing the interesting part of rather given by the long by the prominent notes and their respective time intervals so why are so you have to do is is kind of like a noise reduction in UK because the she recalled something you not usually not in the studio with of the here this this song on the street some all some but with somebody walked out of the radio something like it as they would this song and taking his cellphone and put it next to his record of get of possible every as an idea Dyer some for recognition of something like that I'm said that the lawsuit action by the moving traffic early because of the existence of the PM that is the sort of the new group the samples and overlapping Frank from and you take the different frame sizes these are at the point in time where you record what note that it is at this kind of your you scale of the man the see the melody not as a continuous think that as a seconds off frames to it may be train was 50 minutes at of of the and the strains of that went to get the full amount of lack the different notes and in India and a new chief from the mellowed the though you can individualised every in a pan with kind of head of the idea of a structure that is that you can of to them overlapping so are what you do is basically the goal of the piece of music and to get all the differences and and when the told changes and you have to do it a little bit overlapping because otherwise if you if you do it that way but it might be that the change of tone from just 4 said that they did and the record of the features of the book but that when you do it you you into the 5th and the you doing the 1st frame because nobody knows where to start a pizza piece of music is seeded in quite for example a sell out of the direct aerobic while will allways gave the 1st note and that people get too was the 1st note and and if everybody in the quiet has signed this 1st note of out of the script at the White the right height of light intonation for that the and doing their subsequent notices as it is a lot easier because I'm really doing the intervals is not so bad as just enough like jumping on the right note in the 1st step the more difficult focused ignore
22:08
it and it's kind of like a very good by the when converting it and you take for each frame the spectral sample and you Calculate the pitch from the suspect how do we Calculate the pitches but the below the about that well some offer part follows a basic idea of calculating but typical exam questionable it up but at you boy and you guys with as the idea of page algorithm is tracking the home on except for the you do that you can do it was maximum like the home comic product spectrum all whatever enough that the idea is all ways finding the home on it and see what caused the actual ground told that builds the home on this is that it could not the fund and the frequency of the that the highest frequency of something like that but it's kind of like I'm the frequency that creates the most comics signal yes so we Calculate the pits time that they were on the with the effort amplitude volume it would William and and the more you off the pitches to law canopy determined we just take no of this is that this is no sound at all this is a bright side effect of then we
24:08
find the note boundaries from because between 2 notes that either breaks so the singer just shut up from a nett loss on were less on frequency jumps sell the singer changes from 1 note to some of the best but a shifted wickets between for every frequency jump shoppe jumps but we have Walliams somebody that if the law allowed we will enter a new silent frames so we something like that was some note here and the frequency jumped up and you know and we will introduce up distilled silent frame to just discriminate notes and that because of cause you can do it without without breaking another like keeping up with of time are law firm so that was no noticeable of break in all you can go top of the world a break in the way was interviews a break make things easy of discrimination after with of this really changed lamented the I'm it is the ratio of the success of frequency exceeds a certain press slopes of the than we at a new silent frames of somebody and and goes down by slowly and we see the tone changing but not really appro then we will have to introduce silent friends this is different so these moving to what this is what comes found if we have to framed with the same frequency and reversing the frame in the middle was different frequency the very sure sicknote with that something about the humility the and it's very probable that this was not meant to be that this is just a of mistake may be somebody catching a breath of doing something about it but I hear all want to to to continue the note of the white that this kind of no way where you could do this and this and that it just doesn't make sense and music Fiore it doesn't extremum at the just off the facts of the voice which of breaks of the picture for so what it would I basically do if it is cut out and smooth it over the distinct single frame in the middle of the kind now and that of is what really recognisable as an old note but of very sure break in the frequency that is the frequency of its neighbours that will just moved and some if we have to silent frames and between that I'm the frequency various that little bit bomb that we can replace those frequencies by the by the average frequency self if somebody has silent frame and a silent frame and then nett and of like that in the frequency of that we would smooth the house and next that it so I'd and all that and some singers these so colourations icon of nights and make for a good voice but for recognising the load people on political up the left of we does make that but by now we have to find out we we can move the sicknote kind of put the signal a pop so we can see what of what happened there but the next step will be to connect the pitches to know enough time on the pitches between 2 side frames and note that we introduced over silent frames where they were up for up changes in the frequency so we can say that between to see silent for those allways 1 no almost more and load of the same type which which would be so we do it is we can make all the pitches between silent faint into a note and depending on hominy trading sweet from Act which opened the duration of the of of the sold in the shoppe noticed maybe 5 frames and it 10 frames was the same more than 2 caught a note of 20 than it all of the owner of UK and and just put it to get some then we remove all the note below specified minimum and so it is just no like of the with just headed out it may not be sufficient for a bleak carrying full of of melody and finally after determining the notes the remove all the silent claims because that is what really that it not really all that they do not make melody based on harbour silence friends to make the rhythm they so by removing the silent for I'm changing something and the not in the Melody as it comes in the terms of notes but in the manner the as it comes in the neighbourhood of notes so far go the dot it something totally different from going to the other half of the removing the silent fame in between the 2 different feelings about the we losing some with using rhythm of never less let loose rose from what we have now is the man the with no tight and direct break still silence number of get and now we
30:45
have to kind of represented and under well if we represent of very accurately with all of planes of all the different gradations by by this will lead to a large amount of debt and what it can offer the do if you can use the for much represent the me for would has all the signals although all they are McClymont say when a note said in what the duration of the note for the height of the noted the empty to the William over the No 2 of a kind we can do that we can represented a like of a nice representations but the point is with the help us is it really a big difference in the Melody somebody held a note for just 15 minutes 2nd longer than somebody else it is the 2nd longer definitely its 50 milliseconds off 100 milliseconds problem not so we can definitely was causal resolution which is to to find out the No 2 seed see what they mean rather than I'm than than sticking to the exact some to the except for my to the 6 sampling of what was in put under if you are simple system the rough classification of the melody might with 2 and 1 possibilities to do it actually it cost and now the question is what is a possible the how would you will represent a mellow like that and the and the strike you as being similar to some problems that we already experienced during a this calls of lectures it up at a pet but but it not so no idea how 1st think it interesting yes the UK the while the British for of the is that good in all world like the old exactly only exactly that was the was basically chain coats what we did was which we took the pixels and noted the direct in which the next week's up on a close call to that not doing the same tricky and he would have made 20 is a goal 1 of the of the of the city 1 of the most wonderful decisions and it would not be called puzzles coats now but it would be called off the you it exactly the idea that that works UK at the positives
34:24
code of the virus simple classification of the line was invented about 19 75 and the idea is exactly taking a chain caught up and it was up you move to a higher note was a repeat his stay at the same level of and was down the road on to a low of what you can do this you can buy has also recalled that the set aside is so the into a bold like how far the go down all followed the got and and this is the same this is what we did for the faint code for the chain we we with that but we have different direct that we could work and maybe we don't have to be truly the concrete to adequate for what we actually do it may be help if they left right up downed and and the head of the basic it I'm so this is what I basically due in in Miller the read recognition and like we stop that some pixel with the chain code bills stop some note which is the beginning of the man the of the week to use a special symbol for that and then we go kind
35:38
of like year Beethoven's old to join who can sing the old of joy them so much that I'm followed a should never be telephone grim told for only to use your time for to through no good for but the so it's kind of like the representation of the line and when you do it is to use a well this is the 1st note so you get the 1st symbol then I stay at the same level you say get a repeat the and you go up up and you get up we continue to allow you get the up stay on the same level as the repeat UK and this is how it works so far every change in the note that the record what happened on the relative to the previous APEC is basically a very simple way off of of transcribing doing and
36:47
the and the Ali said it ignored Islamic electricity accepted the rhythm something although preside precise note in tools but it just goes notes from to note and and records of the changes of the advantage of that is that even though my singing Willstrop's of that would be enough to buy the chain coach but you see it in my to recognise the role of going to go up in the Indian note and the in the pitch and so it's records and up at the end of basically you can it also doesn't matter I'm what went noticed objectless because everything is taken relative to the 1st that even if the melodies used the soprano popular like which obviously we come saying my eye maybe you out it doesn't matter because it's all related to the change of policy of takes everything relative to for the 1st noted stop low can go up without going into the final act are so that they kind of good thing part of the reason was ignored which is good thing and the scale is not not that the snow transposition to to make a major part of the mind up up I'm so it's completely atonal which is also good for most
38:22
singers and this will bring us into or the to a having fun with puffins codes and often actually do a lot of work in recognising some simple but some simple pieces of music by the coats and matching the McCanns out what you think of book is still mystifier since he was signed this be activated in the field chemistry the also died film making and the like to play the flute but the what is known for his for this book is written in 19 75 the direct of tunes and music could the and I've actually brought them to the
39:08
book the both want this will be yet so that tools by
39:20
Danny spa sons and what they actually has has done with this book is the 1st index for music so a on fuse idea that it was to describe these each Milzie fees for classical music the visit is it was said by introducing the puzzles his whose including with repeating going up for going not stopping Flamborough from the 1st point and that he has served the world most of the classic says it enough to go up to the 16th of mild and get a feel identification of unique in TV cation off each musical piece so this is what men and that there may be difficult so this is what the West has not on his side but with their classical music bases and he has described them in December called and as you can see the beginning for example of she is quite bit is like any index off any book it's quite see for for this melodies so they blunt differentiate on the 1st 5 an they make a began a case that
40:50
the bigger just that they
40:53
up now so for example the
40:57
1st the 1st and most Langstone differentiate on on 21 the forest BA 15 months where they differentiate on the last 1 and this goes for and based on the frame seeking more and more are on the move back to the size of a deck and change but way is for example of this
41:23
melott each year when the smell of the here we will are in the game here and unique differentiation so index by only out tools of 11th novel and as I've said It's a collection of 4 out of music
41:41
Rossi's has also owns He makes
41:46
the snow and guns and he
41:56
has demonstrated this also and musical notes off the
42:01
top of it but OK
42:09
so this must have been out of work it's a day to go before
42:12
the just was called for a full direct the national
42:16
anthem as he has done the same for each of them has 9 and on the notes and well
42:23
so it was quite well if if they could like the assault actually think became based it cost last week it seemed Mizzy paid yet to a member and it was a service that allows the while dustbowl include use such Boston scalding book it and the goldeneyes the music peace it takes something which is a
42:55
bit more well non them
43:01
the God Save the Queen 15pc was a bigger should have this making we don't want to move anything I just wanted the 5 books yet half the
43:28
85 books it
43:37
up so what do we
43:43
do some quanta's such and
43:47
for God Save the Queen Nihill she picked up of
43:53
all the will be the feat of the don't up of the
44:01
after the defeat on soul
44:04
the and the God of the
44:09
loans up they see
44:28
not so it seems that most of us in the US has been based in his book because seems to of this 1 also but we have also been as like and the more connected kingdom so it can be found and the music they get doesn't would Woodward of clothing
44:47
for domestic Bisons book
44:49
but the public is a 0 0 go away at
44:58
As you remember last week with discussed about creating something like this and it's a billion for the fact is that it has to be time invariants horrify were going for of search for a slump in its likely it's not like it but they would stop at the beginning but with their something that has remained in my mind the problem with Miss about this book is that you need to start at the beginning so it's not that bad but it's nice it it's a great idea it was a world for 19 75 were the joint lead with something like what we Bellamy domain does is not enough not like this ID we can continue with
45:46
the matter for about a week and wily idealist was was kind of a similar on what he would doing and and and and popular songbooks because usually have and index and the and that does not only record the title of a song which very his for popular songs the unknown to people and of some of those but they have no idea and what some is actually called so also finding songs may be a little bit difficult and buried off in songbooks but you have the 1st line of the song the 1st line of the text and the and recognise a summary of with his 1st line up far more efficient than with the with the actual title of some with the correct matter that of some of the basic that the of possible that it was a good idea and still still works in a way I'm so what happens basically is that you get a possible spoke of the period up and you have to compel the fastest was all coast from the database now the problem is if you do the exact match like we just it was was go playful Queen from that's quite easy behind you find similar so what makes assault some but basically if some nerves are not course it may still be similar and a special I if you if you include the created the coming of the singing of something that it might happen but you need to think of says the green but he didn't really and that the tree was system would say what you will run this is not the same as the Queen so icon up you get but the good retrieval system would say while it comes quite close to car says the Queen so eyewitess return go safe credence to the view that basic but if you and what you need for a so called at this time and again the same like would it be full you know like that different ways to edit destroying of up stones and repeat than the cost of the move say could kind of like change 1 or items you used to have a retrieved the read the repeat in in the period but you only have songs that go up or down on the position to you change and and this change or curse of a maiden first some cost and you add up the cost and the more costly this to for 1 straying into another string less Cymbalo all the talk of this is the basic at the end of the problem so it is that we don't all I'm if you if using the reform of the main line of of some some of some sort but it may be very difficult to find what part of the piece of music he actually spent so what you basic the have to do is you have to take this step strains of what you did the and try to make in any part of the face of it you have like the in the database and that of a full stop to stop with the characters and the goes up and down by the end of the of PM and go along for the 3 minutes of 5 minutes 0 of and some of this may be a long and many of the period and the 3 again start here and just goes up down don't repeat up but stops because it was just the 1st 4 from most of the book line of the song for you have to do is you kind of like to Machat against all parts of the of the song because this piece of the men at the might happen in any part of the fun it might be the starting noted might be the refer somewhere in the middle of and unit do not know so you have to kind of like shifted of the song and compared with a get the best
50:12
results of the scan of the interest in part time to glamorous and the mutate music editing distance is basically just a note left out the full snowed somehow makes it way into using the existing note is something wrong so you just exchange vacant something else sever notes are combined to full model note that long loads of fragmented to for sure what happens very off if you are couple of your breath like Maria us who of us can so these are 5 different ways and they have 5 costs assigned to the PM and depending on what you consider to be more serious enough if somebody these out note this might be more serious than somebody fragmenting and out the Nikkei industry of the costs of making the wrong note off augmented so 5 cheaper and making a notice left out more expensive in terms of cost the interesting part after all of this the total cost this is what the editing distances about you wanted strands all your beauty straying into any substring of same land in the fall with minimal costs the trying to minimize some of the costs incurred by the transformation up a good when
52:00
basic to do it has yet to strings notes he knows he does not so early have to be the same number of notes because you can Freckmann notes of which makes more note from 1 knows where you can put notes in these notes out so you have different ways of changing the and the measure off into a bit of discipline that ready the distance between the 2 strained can be given by the end distance 5 operation of needed and you try to find the 2nd with a minimum costs in the UK the I'm Keswick could also do with the comparison open out the and just look at those costs Haapala often do something missing did just in a switch but and you do that it easy but at it the local now for basic me imagined to strings a 1 of the 2 and 3 of their 4 and so on and this is 1 of the songs and the database the and then using the dreary and you stop was was the cock up another Hugo beauty and then go a want 2 if 3 of the so you just as the local up the cup of the and and the and the beginning what happens now is that it's not matching of or if you do it on a note by scale however if you allow for care escape that an Companhia had comparing and companion matching almost perfectly up again which is more sensible of this is the 2nd behind because it happened very off input signal is not perfect this kind of the idea of a white don't of note to
54:40
no doubt while operations that are not really the single and the strength from this kind of cancelling and a week in servicing the characters invented a new notables and there we replace a single character instead of up to go down for example where there is a string of characters by a single characters which is kind of pudding fragmented notes to go the wonder why we replacing the correct about a 2nd of characters which recommends the note into a different different pieces and I'm every character of the strings must be in Boston exactly what operation the weekend up change something and then changed again the but it should have been changed in the correct way before this kind of the and the and then we make cost a book which is basically a bomb if we take the of string while you siliquae restrained and the song from the database of this is the sort band for every piece of Cherie the match tricks cost table records what is needed to and for me into 1 of the pieces in the other strict of it was the cost of to do it it at the same time and the cost 0 that nothing we need to do to change the but it's not the same them we need to find the cheapest operation of making the same and of calls and assigned different costs to the different operation that we a player so this is the
56:40
basic ideas I'd do I have to strings and a half to 20 for them into each of so stop with the 1 strained what can I do to the are icann just leave it for this kind of across out like and transform it into something that might need for the 2nd straight less no need to transform it into a non or something because icon of need that at this point that I'll need you for this transformation of the characters by could also be in certain note before it and stick with the rest of the this is basically the this is transformation and this this insertion but again there different things like into and of the fragmented and consulate dated and block of a pet but word from that have to live with a new strained so what they can to for example for this drink comparing the new strained 2 of the old strained if they defeated the 1st 1 last have to look at the 1st a possible entries of the strings and perfect they do match though prop the and under FFT with the remaining Street well if I'd just introduces just confirmed after look at how it works well for the 2nd the 1st almost 20 points that it had of the 2nd whom Beso matches perfectly after go to defer from so ought to a needed the well you'd be doesn't help me because they need to use in the beginning in session or by could just the PM changes to the of this is the insertion this is the change the transformation a press Apple with a different ways of changing the Stream such that it find resembles the string but along to the right of and the different ways well through this treaty in during the different costs and the 1 way this is the minimum costs defines the distance between the 2 straight if it's easy to transfer for my me into each other the mercy of it's very hot to that of the 2 could very cost me to do that than the December Kim 1 question remains how do you do with it because going through all the different ways to find the minimum costs heard so why that no 1 would dispute that this would be 1 of the cost of a teaching last would be a year of nothing but we did here was a transformation the cost of the transformation of what we did here was also transformation again the cost of the transformation and that it was the correct thing and so this is kind of the cost for this part of the point on the other part we might have what the look at this part of the year before 1st the transformation so we for the cost of transformation from that we have no cost than we at no cost the and the way up the cost for the show of write either of them Commey small bigger depending on the hull big 1 of this that the smaller because the assessed to time seedy and longtime see the and this is only 1 time city and to the so this is a name possible costs 1st this might be fined the distant web of them like the also other possibilities to find to find the cause minimum to see a cat would be to be a try them for you try every step self I'm transformations you could do to a string to get tools some of the guests although the all Capkovic in this part of it can't tell that it's only the idea was to do only those whom transformations chief and and stop the cost of the high of the some which is called agreed strategy the Oasis go for the cheapest transformation and had on 2 legs transformation of the necessarily that the problems with Green strategies this the expensive Yale in 1 hand while the quite what she because deciding only for the local the cheapest transformation applying at and looking for the next transformation but businesses is kind of kind of an easy way to figure out which way to the very complex Graphia is needed but the problem is of calls that might not end up with a minimum costs there might end up in the local minimum so there might be a way way incurring more cost at the beginning of the transformation was save you a lot of costs in the end of the transfer mesh then agreed strategy with the wrong and as 1 way to do it very efficient
1:04:21
the bomb and in the matching and that the programme would come to the them and so I'm basically from what you do is you apply the operations from left to right so you start with the 1st 1st 1st notes in out and I'm what was basically done is I'm you have the notes that result from the prefix of the and the Suffolk of so you work your way to walk through 8 and where it would have to consider the Suffolk was still to come because that still has to be said for that was the 1st half was a prefix of and was be it it the other way around so you you you kind of like a looked at the at the note that you need to be to build on that you still down with a certain part of it and I still have to do the rest of of that kind of like it you want to transfer for I'm the suffix of a into was the critics of the prefixes over 40 created and then you go that means that in the graphs for the transformation you have N NYTimes and to because you can't France for me any of these M erected in any of the beast in the depending or you do quite quite a bit of a no what you can do still spent on the NHS than the and time graph but if we do it prefix suffix wise it works at away from but still and all of stuff of stuff to
1:06:25
go through bomb since the costs on nonnegative so we just assuming that no operations no change comes at a time across smaller than 0 so but it do you know no way of adding something to something that in the best case that distance between 2 songs 0 it's never negative time then you can find the cost by Dynamic Programming which is kind of a very well known based paradigm in incomplete assign so that this can who has stumbled across the programming so what I guess due in algorithms the structures the typical way of doing Dynamic Programming 1 of the major principles of computer side of of algorithms and dynamic Programming that a way of finding the minimal cost without having to walk through or the part the problem was
1:07:40
how does it do that for you you have to start matching the different notes and the new can kind of may be recommend a note into different notes enough or you make one on one comparisons might happen and but the point is really that
1:07:59
some you have different costs for the different for different actions that it to sell whom if you replaced something you just have to sing signal and put it together that 2 1 out the noticed just Freckmann on the same frequency as this is the very use of error should and her very little costs if you change but don't to don't up that's not the usual cost of that should be a high cost assigned to let because that told the changes familiarity the if go is something to be different and if though that got it is a totally different feeding so the cost of that the behind inserting repeats of cheaper than inserting of ups and downs because the kind of like keeping a load noted that longer is is is bad of and then changing the frequency of baby half the cost of if you do replacements the cost of the operation as a repeat goes down so you can apply the you thought it was the same note but actually going down should be smaller than that of the euro dollar the mellowed egos up so you can find a way to that and signed different costs of this kind of
1:09:37
whom saying it I'm also the you in the ship the treated in the same replacement it doesn't Madaschi replace something and the dollar you up from their in sessions and conditions that cost about the same time and similar cost for fragmentation complementation and equivalent in search of what I do should be kind of the match but meant something over the but to go if you don't have a are strong reason to believe that some mistake might be more well more difficult or more alienating from the from the actual Mavity than the other than he should and the and the same and the state UK but if you see what this is really Imeem in this is not something that happens what seemed the input line this is something that was meant to be closes purposefully the like that and you should basically assign higher costs if that has to be changed from the fiddling with the
1:10:52
system that but not Lescott down to Dynamic Programming basically the idea of dynamic programme is to consider all the different notes in the graph what could be transferred into what are and it doesn't matter how high the associated at a cost of unlike in greedy algorithm because spending that in the beginning might say if you something in the end so it's the same as in life you know if you if you buy through a truck for with a high quality it might in the end safety some money because it doesn't break that the for example of the by cheap for says you money in the beginning but then offer you might have to buy a new car but because it broke down and the and the launch he was just a part of something of the same idea as it is done with dynamic program to consider all possible notes and the craft nomad a whole lot associated at a cost of that of course accused of Optiver entirely from the results so if you see that the community and the results have nothing to do with each other the and it's easy 2 conjecture that a transformation what with the result and to a high cost and you just ignore the notes and the graph which up only reachable at at at at a very high cost you begin all the notes from the which can be reached only by the conjecturing is when and the you have that note in the graphs the total cost will be much higher than of should be expected this is the kind of and of the idea that we need for and so we use the match fixed of the notes of a and B itself on 1 side of the match expressed a I on the other side of the metric is the and each no to look the group was the metric is kind of the cost at what cost to get from this suffix of big so the respecting prefix of the the kind of what you have to do to camps for the rest of the into 1 of the but biopic was just justified
1:13:46
but for so the Eighties of you and the mental strength and the bees by a pet and only examined the 2nd that at inside the window over here which is called slope straight because what that means the kind of yet but the idea is that when reached this point the a the string has been converted into the and time the strain that is a have to make my way through this proves metrics that make it more clear for part of the so this point over here it is basically where you have the origin of a strained and the origin of the this point although he mocked the point where you converted the a string into the straight but if I'm here the by convert the 1st sign a 1 of the a string into the 1st sign of the strength a the knack could go here and so it well and dealing with a to but I'm not using up the to which basically means by just the way too so moving into the direct and the moving into that direct since is although insurgents are because of the surge in part of the Beast drink I'm not changing was left of the strength right moving in this direction means it up but but it if this is a transformation exactly because of what you sign from the a strained is transferred into 1 side of the street into the correct time from the book care and Niall but or what these transitions means like it also say that every possible in this graph if transforming the a stream a string into the beast think it stops Olivia and somehow makes it 20th Independent whether it goes vertically or zone to lead the side it uses a different set of transformation to do it like that so why could tools for example is a good 1st died of their after string 8 and it seems global after note of the a Case but like could do the other things that could and so while the notes for the and Latin died reais obviously this is a way to France family 18 to beat the of using this is very very bad way to think that because the cost of very or up the different moves different costs and so what I'm someone of trying to do is I'm trying to reject the sequences found with with was was higher costs and I'm not sure where to go because and all that and you'd to end up in this at all the ending of the measles edge and have a go back it up so could do something about them going down here than going up here at the end on the could command would be sent to bow to do so 1 or 2 costs is increasing exactly by this flew over here because if I'd have gone directly down here which are often as he can do what it says this cost so this new way of sneaking that the but close SIAM to the girls the mood of the work done and undoing some of the work will only create more costs for this cheaper way of coming to where you want to go without a this is the basic ideas by by by introducing these the so called Slocum constraints and and just saying about I'm at you it has to be the Bagenal enough but if you moving to far although some Khalid off and something as from and there is a way of doing it with the cost of these other slow constraints which can be computed them by some tricks and the and it's kind of easy to look only at the remaining window and the possibilities of the remaining window and crude everything that from the start what kind of like but it remove the following encouraged to a high cost so if you go down your for example and just been into everything that is the end remove everything that 8 it did it would be like a single belt would be more such that enough when he was enough he would insert the complete clad not with the before the and and adult including of not a sensible way to do that for me if obviously to a high cost up at the stadium the window and this is what Dynamic programming and it would lead you to a guide you to the best possible without exploring all of them brokers very good and very nice principle found that should have been coveted now doesn't actually and that is definitely interesting to know about
1:21:15
cell that 1 problem was still that with the kind of like had to match substrates so a strong a as usually longer than the period because the only thing that referee in the 1st 2 of the 4 1st cup of notes and since it may be the referee of from some some some phrase of the melody that is not the beginning of what you would have to do is kind of like you take the very long sound and the rather shalt period and you have to move it as a sliding window all the salt so the Treecode Perahia politiker Huda or might of the year and it might occur in a time so you have to be to do that the Linear Programming competitors shifts for those substrings of the quite difficult than and not off work do it well
1:22:26
that's kind of what he at would have to do and they should take a bit of a shock when we do the different groups to go up at time so it is a lot work of but it can help you to find the melodies found weekend or even to the same tricks like would in the in the chain codes and that they were found and I'm not interested in the I'm in in the strength of the pitch change though the death of to effects on the magnitude of that high could say no by M not only going up or down but I'm going up and up to point going up for a battle something and kind of raclette the number of semi tells the interval between the of the note which basically results and in a finite granular M composition and the cost code on which is called difference code of conduct
1:23:33
also recording different and and you would kind of like to of the passes good year enough votes to join so you repeat you go out you go out and about what the difference part of the difference code with to is you repeat basically 0 you go up 1 step you up 2 steps then you go down again under the repeat of and then you go down again to steps and this is kind of recording the into of up between the 2 of us but of credit could be also bigger jumps so for example here if you go down their that the minus 5 and if you that impression of the some of the of the melody part of a feud with a matching and the cost is codes now with the difference coats what do you have to do at at this the number yes the but that well that is that 2 of the as they are mean time but but but you you considering all the matching not with Boston to cope with different timing you can or a with for the difference code into a possible because the limited from the nation you can do it the other way around because I don't know how big the intervals will when I was just saying up and down the what would have consider when matching those codes those difference codes unlike what you did and matching the possible but as part of that of the ball so what yes exactly so you have to take the differences the absolute sightings of the differences and put it into the cost of the measure if you have a very small way all jump between to know it doesn't count for much if you have a very big job called for a not changing if you really have to do 2 to make up I'm and the and the and the transformation but has a big into about it should cost much more than making up the transformation of the old because the small into because that can reflect error and and the and the singing of the with of ability but the changes are usually purposeful but was that basic to what you
1:26:40
do and the interesting thing is if you look at the distribution of them into a bowl stump in in music databases and here Kasuri M has taken a 10th of songs and and look at what other prominent intervals it it's kind of interesting to see the cell of cost you have a lot of repeat and that's kind of Boston distribution so that it sits its less probable that it will be only a half tells because this is kind of like a in Indian music I'm hurt as as being under the and that it had offered the I'm very often it's just told and just 1 1 step that you have the 1st cautions wins this was kind of interesting in both direct intellect symmetrical because the same steps you got used to it your that out of this kind of had a kind of interesting to try to a which was of this consider a kind of a painful of by most people of because of the size of a 7th side of the interval at the kind of strikes sounding funny and this was 1 of the until mould of used very off to focus of his listeners and because it was very unusual to use of the music and we can see that only from also from the database it still very and to to use of but in the end it's kind of like the you you have little 0 Big steps you have many small steps off with a caution distribution of a pet interesting to so the
1:28:33
I'm advantage of different goes up busy that he of precise distinction concerning also the size of the jump in the world and you can wait until the end of distance costs for because jumps on higher than costs for small jumps 's advantage yet more effort and matching and other more accurate segmentation that he seemed to see that somebody went up in the note is easier and to see how by by about how much he edged up go up and is also problematic for the full acute because highly of you can sing a proper quote it's difficult not isn't and and you might end up with a total with with the wind and the and the system would not be happy with you so maybe it's easy to say you just went up for the loss of some and so it might not have the effect you do you supposed that has a because I'm most pieces and music of focus on the details rather than of the big ones book but this
1:29:54
basic your want to say and now we can break the say Well 10 minute break the book
1:30:07
Bahgat ahead MEPs belonged something that we would discussing about right now walls of kind of finding the Paulson's cold finding a discrete representation of the the and then matching the Miller the again to the movies and the UK what we could also do is we could use a frame based
1:30:36
red look a frame based
1:30:39
representations not segmenting the individual notes but reflecting the called to of of so the idea that we sampled the tree and then
1:30:55
you get to walks which frame a representation of the pitch the of recognised and how it changed over time and discount to a is the interesting thing that we consider to be representative of the man he had not edit disturbed was of symbolic representation of ability the but from time series curves how do we match time series against each other how do you describe the common to lose of man of the interesting as the such an ice idea and when a comes to matching hmmm a the every under it so like fact that 41 possibilities to to do what to do with the real matching but will not discourage same area so maybe we can be used tricks like that load of a features like that of the area of different different types of on to statistical and other the of to exclude something from match so if I'd do would have occurred like that just looking at the area Cansei can never be a good match for the public until the at the idea that here of so maybe as the rough representation and and excluding some parts of the database from being considers that is a good idea that all the actual matching of the current they have to do is another trick another 2nd and that it was we basically the have to compel the points in the script and of course of the I'm was set before in the point to point and notes to note comparison doesn't work properly because it might be shifted it might be some all some distraught at the same applied here
1:33:38
so it's it's kind of an of kind of difficult to say I'm and we will in the moments he 1 of the good possibilities to still to the matching and the and the time so I'm ready basic you do is to take a framed classification was that they could distant sampling said the advantage of this the is you don't have it currencies by the direct segmentation because once you decided for for new pitch to be a certain note it will always be this note and if you made a mistake I the mistake and the base which is that with the frame sequence's said that it's easier to some of the the well to do to measure lose their and and give a correct representation a 2nd possible that 2nd 2nd advantage is that the famous Perkins's also contain the rhythm inflammation because you see how the pitch developed over time and that is something that is hugely which account the with possible but the retrieval times are significantly
1:34:50
higher because of what I have to do you have to compel time series of pitch families the change of pitches over time for most of the men the and the point rise comparison as lead to pull results because the speed may be a slightly different enough you might think something of a little bit quicker so it's still the same sort but you just don't have the right Temple you just don't have the right rhythm which would definitely change the frame reputation for 40 to new matching will become for from the left of the
1:35:30
problem and would be due to a problem with this kind of like a dynamic matching where we stretch part of the Miller the that will some rather quickly or other slowly may be complicated passages of the melody or some rather slowly in the period to discriminated that a or 2 to make the breathing even easier and a and with the best of this so called Dynamic time walking which is a method from from from that a mining are which has the same principle as edited students it trying to find the minimum distance between points to 3 at the distant point on one time series on 1 close to point on the other Kerr and I'm you don't have an alphabet that he after kind of kind of transforming to each other but you just measuring distances between point of Kent and the higher the distance the highest the total cost of the and for because in the end you just some up all the distances and that is
1:36:46
that is the basic sorely basically do is to 1 time series here you you in the could the point and the different times series here he here I'm and you measure the distance but for year and what I basically do it is you measure the distance between points and since it should be dynamic and should be a point point comparison you can set about this of or so the measured with respect to this point because it's kind of the fragmentation of so this Norristown here actually to notes down here because it was sunk and in the origins so it's a little bit slow of 5 at the end you have to arrive at that point in time and that is that the sort what you can do it is you can speed up basically 1 piece of the Miller the against the other will you can slow down 1 piece of commodity it and this is what makes it Dynamic you don't move was sustained speed but you don't pointtopoint comparisons but me regulate the speed of each signal the suspect to the other so sometimes more notes of 1 signalled or compared to 1 of the other 6 were sometimes 1 note of 1 signalled compared to many notes of it in the end you have to and at the end of to if to compute compared to complete time series but in between you can stretch you can move with the time and this is what caused time walking up at but the basic idea
1:38:58
of what you do with it is exactly the same like would the editing just enough to time series 1 down here UK and and the 1 0 up here and other said he the F to measure the distance so he of the possibilities 2 men were up with huge black again to measure the distance again to different points for the 1st point it's clear they have to stop the same thing the 30th but not for the 2nd point you can't you the still compared to the 1st point that means letting the speed of the record OK the you could compared to the 2nd point to the 2nd point here which means Green the same speed of the L you could say that you want to compared the 1st point of the record for the 2nd point of the Blue Cross so we think the speed with which it traverse Cross of care with a 3 possibilities to get the new measure meant for for you across and at the end have to arrive at the end of the 1st and of cost you may never go back 1 to compel point of the game the only moved into the direct of time of it yes she the across the board on the wall for all new ones but yes but the thought of that yet all the while the move will be up for Welsh how would you do that the but you can not stop with the 2nd point yet to stop with the 1st point the the while yes had in order and the UK pulled it off food prior to that it but of the 0 the point is you don't remember anything that but you just moving along in time UK and a new making a point by point but for each point you have to decide what to comparative gets you can of competitive against what you already compared on the record you can of comparative begins was in the future of the red rose but he for every point on the Blue Cross was basically 3 choices and this is what you do he you have to do the distance for the total some of the yes to compel the 1st to for the next 2 you can do exactly what the what the Baltic line safe enough you 1 1 pm you want to the 1 3 basically the 1st 1 is basically and in search the 2nd 1 was basically the the and the 3rd 1 is basically the transformation the issue would save him at the time but you have to decide for either 1 so you have to take you want buy it was by being 1 2 Street UK and and then he went on to the next point on the cover from the and say it was a possibility why can't you the depending on would be before maybe I'd be cited for doing it like that and for the record the of life again of the possibilities of state moving there was letting the other across UK and at well with the possibility letting the blueprint letting the record that going pointbypoint so for every point of the curve on this possibilities and is exactly the same that we did in the in the editing distant was ordered the dynamic Programming problem of for the Games but the transformation steps are so we have the points on the record with the points on the Blue Cross and we have to transfer the record into the Blue Cross of care why can't let the Blue could icons where the red card icann to point to point comparisons these are the 3 possibilities the and they reflect the Sri possibilities he the Boston life for in the end for the total some has to decide what you want and that reflects on why can't tools for the next 4 because have to end up here and again iconic 2 circles a can of to of this time for local leaders who you and billions of the world but you have to go more 23 into the by rich of the of and point and you can never go back to this nope but UK air understood same problem like before same solution like before again you can do the dynamic you can do the dynamic Programming on the script find the minimum costs and the minimum costs were determined to cost you the 2 of them but at how to an know what this step costs so what is the cost of this is that costs because the those 3 possibilities that F a but put it interesting question and yet as why is it called pandemic time but it but it up but it it but is the because of the costs down here are not fixed it's not like we did with the Paulson's code into a bowl with cost not to know what transformational ois costs from all of that but had a big the cost of fighting Kurt of arose see what you to discrete basically dependence on you well on what you experience yet because she 1 you to see 3 reflects the distances that you have to add up in the end he won 2 match should such the told the distance between the growth is minimize look at so looking Mariella clear this felt so we conceded that are you and look at this point in time to stop the possibilities committed against this point of the possibilities of match at this point and possibilities match at this point and this point basically a case so I'm down with this point Dalmia I'm moving for the next point full pink the curfew the
1:49:01
I'm staying with the point the walking He 1 week both curse this 1 earlier and the costs of different because this would be basically I'm move the is the key coast of Kent so what is to see what you see what it is and 1 step on the path of this is that the PM at the Red at his time of moving 1 step when before for example at this now before but all this from a distance the of moving a step on the F means exactly going from here to here a moving a step on the G Kerr means going from a to B and moving a stepped on both curves means making the decision of this 1 as well as this 1 of the and what pinkoes these costs will obviously this year and this year is quite low and costs 1st this year the huge distance and would give a huge effective for my total some so maybe this is a bad idea maybe these 2 was a good idea for about the cost of directly associated to the decisions that have to face of my costs from the start to be game opec evidence of how it works Eichenwald the relative time of across of F Eichenwald the religious time off G at the end the time has to be synchronised again at the beginning the time a synchronised and how there the dramatic but to have to get there with the cost and cost or the endemically assigned depending on the distances between the 1st about her up is
1:51:46
the basic would Asiad and and the in Saudi to set up what happens is basically mollified as a team so you really have to move their stay at the same point you can never go back academic 0 now on to competitor the point way down the right of the work began compared to the same point again meaning walking the other current that to the next point meeting microbes full moving both curves which kind of synchronised of the of the continuity you can not make steps to the bigger than what it and other just give that by and then we will come back to the next 1 that doesn't work you have to compared every point at the start you do have boundaries the soul the stock has to be the same the and has to be the same in the middle of you can make it will cost of the but in the end you have to be synchronised and in the beginning you are interest and the and again like before you have off to transfer for more yet compel M points on the 1 growth was and point only other 1st was Dynamic Programming other applauded No and size of the Earth and if you special cases you can make it
1:53:20
up and threw show shed examples where you can see what happens locate so
1:53:29
for example a book
1:53:31
the speech reviews
1:53:38
but last week for the French a between music and speech to remember it's Hugo originally that hype and P P seemingly account for what he could Club the and PP the the and the and the UK so now I've whom modified it efficient and I've thought he just the speed in some places the the and their own way in the world you you need to yes or after us for new debate now we 37am forget that you can already established that and so is the same sort some of the GRAPE opposed to the people is the memory of the time working and see how 1 can transfer transform the modified singer the more defied the sound into the regional 1 in the whole Dynamic Programming about about an applied the pandemic Programming offline seems Sunday's computed takes a bit more time than expected
1:55:22
and that this is what comes out of high
1:55:27
didn't use the typical nation McCrady's might be the stage for a simpler of a dividing the stigma into windows and each window Unionist pictogram I've created the every 2 sold his leg the only simply and the cost of creating clear these every 2 frequency on spectrogram in time windows than any Programming actually does it as as we were discussed and do not always compared to the cost between working 1 c or the other signalled and what you see here is the cost of transferring the meeting cost of try scoring the slow he would the all here Michael now a all injury so on this access saw on the exact is about is the slow signalled and on the other hand is the scene wanted transform the boys University of and I'm working with the beginning of the last email is expected then I'm performing summating where the seamers side to do the same the end it follows a game of the tournament make make its smaller you
1:56:47
can see more another would be secrecy here condo but with the board game so the 1st loping Signor sequence and the 2nd would be the sequence he found and after this would be the benefit of low programming is actually a local copy this diagonal from NEMO cost between transforming the 1 into the other and the way of the city and the Bulls or estates my perception so at the beginning of the season was over time working in the sea Milan mating then where it would be matches them be the and and that it was quite well with its go back to the picture of
1:57:43
get so we see that
1:57:45
Dynamic time or pink automatically does
1:57:48
what we would be kind of wanted to do so it determined the best way the minimal cost up between the but some of
1:57:58
the we of the uniformed time walking distance between 2 time series ex and wife is than defined as but squalor some of the differences in way that your voice take the best walking distance so you both time excesses Ike's tended to and times and so that this a book with the signals have the same length of the from end of for the speed if you have variables speech to speed up some public 1 and slow down the of a possible of of her the for the religious time and that is basically what was happening the
1:58:49
bomb the idea of of uniform time walking is the blipping cost should be basically the diet of so since you since into stretch the signal so they have the same length of moving and the joke is kind of the way to go from that you can also Calculate the beautiful time walking distance of from from different length if you use time scaling of a new kind of scale 1 of you kind of kind of scale 1 of signal the south that matches the other 1 is basically
1:59:34
what happened in I'm a 3rd to get to get a bad tuition for for for you that kind of like the idea of local Dynamic time walking found that you can say about idea and want to stay as close to the diet of I'm as possible but only want to diverge from the Bachelot in small pieces of sound so it's not possible to say that 1 from the the half the time and then speeded up so in the end they will meet again but that but only in in some areas and the basic ideas behind that is if you have to start from here and I have to come get here somewhere than you the perfect way would be to the diet of UK it is owned by Wall those of the best cost though so might help and what you do with them is to stay as close and in some positions you might of meander around the diet to go you like something out and these are the local timeframes where you kind of loyal all for some time walking and you have to keep that up to 2 million so you can pay a basically point wise to law little letting and evolved they and say OK icann Eichenwald 2nd but after synchronised at some step on the head the UK and the US may not take for too long the basically the size of the into is you look at the time and after the interval you have to be back on that makes the of the PM the usual said that the Uman perception but that because if you listen to the speech of 4 of don't with the most likely to Stoltidis it didn't silent for you as perceptions exactly like the other wanted did it because it was really hot to understand the 1st part of it was so slow across so that sold Celso of wealth they to such a degree that the perception became tainted and this is what what he would want to avoid UK and this is the idea of local time yet
2:02:14
basically your with have to stay in the Strip around the badger you make me and a little bit but I'm as soon as you go here from because the and he was the signal for much of the and some of them come back to that but local Dynamic
2:02:34
time working from and the interesting thing is that for for far was lower than and time welcome you can actually built a effective indexes and and we will see in the end of the lecture we do that the indexing part the Gemini approach bomb which basically does a a little bit of what you were proposing that at the beginning but I look at some quick and dirty measures for example of the of the of area below the Curragh for something like that I'm to prove some objects from being hunted at at all of us to win as the air becomes to be you not following the way and more that can be so
2:03:26
I'm what I can do is you can take the Beatles song or something like that you know and then you have the music pieces that based which basically of the club to a off the pitch a Kent and then you do that period by handing out and humming this hugely slightly disappointed so you can see that it's kind of like little bit difficult India and then it stops for the breathing popsy enough any after Competa a basically what I do is you have to move it to a place freezing a lot of interested in and then have to compared with the other signalled basically you do it but you did before the other 1 was of the here built the match a try to get from the point of this point somehow of Dutch will pick up the cost because the differences between the and that system the to mesh up the but after the
2:04:41
transformation to if you if you do that you might find that this is the humming time series of and the other 1 is the most serious likely to lose take from base of the and now we stopped the walking processes for some of the year and then go here whatever incurred the minimum costs but yes 4 all the time that those who in the Orange after the same out again you have to look at such part of the peace and if you don't know which part of the men was the only child to do it is by moving a sliding Winmill the size all the complete piece of whelping it into such a way that the man the of the and tie up piece of music it obviously not sensible but that was the mood of the law might be yes you have to allow for little Luella of Malta but that shouldn't be too to bat of the the last thing on to
2:06:22
due to fall to date is a kind of a folk of on a probabilistic model of modeling the sound of this is that could be represented of all the sound could be compared to the other side and so we do it on the example of the of the state events or if you have the frame based Methods basic you have to be paid the of the audio signal but we have no idea what this behaviour actually is about to go on like this was that the music line of a Jude just submitted ago you now and we don't care where the day doodle whatever it it's just that the band that of that has this contact to restrict line this correct eristic kind to attract like in the case of the of the image retrieval where he said that it's just a point rather shaped by don't care but and a point elephant all way flawed but if it does as the direct eristic shape this for what we do in here and the question is how do we determined such you in in in some audio signal and them in the idea is basically to use a mock of model sold a probability of random pro system are friend approaches and a tight to see what actually happened and then you can have a more cope given by the mock of more that probably cost the so you can have a model the Reduce resulting in a shoot symbolist talks like we did in the 1st and the and the random feud case of all for the image to the basis for a set of well basically you could have a palm generative model that their way is results and strike of making it for all it create something that was at the shape we can do he also so the load of a basic probabilistic model of that creates Beatles files and my with a certain probability that will be made you demanded and that the amount it you know like that it said it like the Aids writing and that enough yes if you take a lot of pace and give them a lot of times to act on from on typewriters you know some of you might produced and at the end again it may be of is are part of the models
2:09:03
that kind of like the the idea behind sale out what we are doing is kind of a consider for example that we are interested in modelling individual notes and we are said Well basically of a way of doing that is be at the attack but where the pitch arises drastically and that the case at some point and the sustained of some some area and and and then the discrete Eastern and and fatality enough and cost us can also happened here in all time series of her so we have to decide he now like 1 of the few really move the total not while the just the attack of the same told again most something like that and to distinguish between that we could be could you employe of populist more of of classical also be Paul ballistic for the whole Timeline of says it all like everything that sounds of that sounds like a Jude could be created by the home of the US led that stick to the case but we have this specific model because we know that this is Holloway note should be if you want to but he on a piano this is how it is and that is human this is all a generative model for new notes but the
2:10:28
implementation of the 2nd of the band on for what happened before the attack the decay the sustained the read the it is basically a string the attack with the case the sustained the reuse and them the sight APEC over some fixed of and these are basically speeds of the model of what because the cost to give and a single note the songs that like a silent like a joint is justice echoes of atomic events but give you a possibility over time to create the result APEC so yes set of states and have the state transition which up part of your model for for example the attacks system that always comes before the cases of the decay or with comes before sustained but it might be that there was no really upset about it but but it was just to the right level so the attack did not overshoot there is no decay so what happens here is to have a direct cost from a tactic to sustain a and giving your morale you can also buy credibility owner of the wealth of usually there is a detailed phase so was a point behind this happens was point when this Scot opens the probability that any of it happens with of cost 1 something has to happen but we some we see what it is that this kind of model and this is what we call a mock of chain of looks like
2:12:41
that you know what we put for every point in time we could stay in the attacks of this is still the tax faces still the attack faces still eutectic that would moving to the case that this is still the case face now we moving to block the sort of until the whole thing is finished and in the end there was site and if the homogeneous Moccas proceso if there was no random this in what can happen but we know or other possible states that can happen than the Irish growing edges for the state have to add up to the 2 1 has to beat sensible to go either 1 of these at nothing on for and the and the tradition probabilities those of transition probability that time in very and it doesn't matter when they have put it doesn't get more than probable along with the attack last that you go into the decay but the probability is just taking a wealth of everything this time invariants makes it mock of so it does not depend on the time it just depends on the previous day and it has so does not depend on something that happened here but really on the previous day only the very local notion we had that already when we did the condition rent of the the because the condition a random feuds we said on a like it's also mock off problem because the Koloi intensity of 1 Pixo is rather used by the surroundings of the fixed what by the cult of the of the of the of the also this lot of characteristic same here we use just the same way that so many of need as a
2:14:50
stops distribution for each note that we kind of like the state's at take for example and you have to get into that state some of dilemmas a probability that the notes stops was in the tax take not maybe it's all point may be it stops implicated in 20 per cent of the cases all and and sustained a set for 1st of the Soviet than the of book this is the in the so stops distribution ineffable ability usually it up but it's kind of a the notified pipe and we could say for example but don't always stops was attacked and never stops was any but and Spain to that of the other in a and never stops with anything else a Case
2:16:00
yes of we have to buy you which is why I went to the and now we can have a different sequence's of the bands that happen so for example of if we have the second the 1st time frame is attacked the 2nd decay than the to sustained by claims and then there is 1 of release time frame and find weekend easy to compute what the probability of this happening before spec 2 or more of for we stopped was attacked those 1 point he wrote of it from the attack we direct the move to the time for 1 3 them from the decay we go to the sustained of U again this is open to 6 this would also be mild 1 some than we stayed in the same never take this view of the Texas to take this route and that's just multiplying although abilities over the cost makes it easy for us to discriminate what the total pro ability of this happening in same girls of for a fee of different so we stayed in the case for quite a long time to say this is not probable of so we have enough of this time 3 times a case for the cost the corresponding cost for this would be a case a go into the thing I'd go to the adjectives 1 time to Times 3 times for the soft triple the probability of that happening from by move to the S to sustain of a move to the release level and hold on to the site in UK at different parts of different probability but we can't get the latest book get up from of the
2:18:21
attack also cost to give evidence is the but difficult because if we have a model and say this model can produce page Woodsong of individual notes that followed attack decayed sustained reduced family from then we can Calculate the probability of something that is so that his generated by this model that what we have is kind of we have we have the stigma we use something we of something that sounds like a Jude Law but that sounds like a typical envelope of note but we have to guess what the actual Monroe its from we don't go in state of the model we currently up all we just in the 1st few load of the 2 would or is loose bold to generally of bitterness limited will change played from and the different possibilities is still at that is already the case that depends on physically on the on on the probability of being an attack are moving on to the Cape and this is what called in much of what because we never know we would never can be quite sure in part of the model which we all we know that at the beginning we are in the with a stop improbabilities in some state and at the end we are in some and but between the 2 that from probabilities 4 different chains of the vengeful different seconds of states happening a Case this is the basic idea behind it so we
2:20:23
say what we have are also nations help the Vega served a certain old out of a certain much in the frame point and the probability is that in the certain state by make a certain or of patient or so but before saipem well if I'm moved from the tax to be paid then something should happen that the envelope turned down and the wind up something turning down this transition is quite probable when the of something going up it seems to the attack because the only part of the envelope with something of up so being an attack is quite possible if I'd make it off the bench being in or out of the other states as quite not as part of this is kind of what you call activation of ability so from all possibilities of a from all the possibility them from the all possible states that he could be in you have a certain probability that he will make exactly that observation to the but then we
2:21:56
might have an observation change for 3 0 5 0 0 1 and now we have to query from the probability that the morale of the state's 80 died there St fixed this operation then there all 3 0 5 0 1 we have the the observation probability of serving allsuite in 1 in 8 at the serving all 3 in the opposite the Austrian at the same goes for of high observing 0 5 today of serving 0 5 in the absurdities of life and the same goes for post but these are books of a sum of the disease for Kent the Zell transition Paul but it but a Case transitions between state of the variations made in circumspect and this is the initial of stopped in the state of study basically do with this well if you want to know it was a medal this was spent Seibel for creating for the observation generated the of the vacation then I have to say well what caused the stopping probability that stop today and 8 anyway Kent and less a probability that being so this is also that being in a file Bazoft 0 3 and was the probability but from the I'm with which to beat from the book and it was than the probability that in the eyes for far was a probability that then I'm moved from the 2 S and and has made the operation or 1 of his and if I'd just multiplied like at the top of probability that 80 S is the Genoa to model the Genoa 2 seconds off the event resulting in the observation also 3 0 5 0 0 1 oka killed questioned it of convinced but so far more than in any case we just assuming to whilst this 2nd of state of with just to get his wife close it might of because you never know the correct 2nd thoughts that we will never know what happened exact but assuming that the sequence with the high his probability happens has never heard for this is what we get to the with looking for the 2nd with the highest of ability and with you this is the truth this is of a pet goods that means we
2:26:03
know seconds off at the base and this would be have been all frame based signal that might be the song that sounds like a Jude might be an individual note that follows the tactic a lover Patent and below patterns everyone to no P is what is the most likely seconds of states and order a mock of model at the time this or that looking for and the question of whether it is possible to sign the sex went all of the waste to the point where all ability it is just well take the most probable seconds of happens and take this has been true quest
2:27:04
sometimes basically is you have hit states you have transition probabilities between state you have the entry probability into the state's and you have the of the patient abilities offers results but and he would sign the seconds alterations to the event whose a mock of model has created the observations as high as but of the
2:27:45
probability that the eventual or Kirch it is already known that need and he has occurred is given by condition appalled but the probability of a under the has already of this book the vertical of find bomb so when we go to do this we going to say what they see the need is everything that has happened up to now and I've a condition that everything went up to now was the probability that the rest happens this is my idea of condition appalled but some things have happened so I've heard a couple of notes all I've seen and and attack the signal was a probability that now that the take up was a probability that the sun would just go on like to of all the same question and we know from of from Paul but fury that this condition of the PM at this condition appalled at the basic the means the probability of those the operating divided by the probability that beat crude at the club if be used the next a possible he it will be very hot for a because 80 0 cursory off than it canopy of causal bump seconds that's kind of the and the and the and that other is the fall the probability density of random variable safety of to random variables which is basically what your throat on now and what your going to authors and the future of its just say that can't
2:29:55
stop Kostic predecessors up sequences of random variable for each point in time this is basically the Timeline something happens it will ever happened has been quest by something battle this is the basic ideas and them and and in my or quest processes this is what put a mock off in the stock plastic processes is a well basically if you are a certain point of time the probability does not depend on everything that happened before for but it just depends on what happens directly before in the step before so at any point in time my a piece of music could start Toussaint like a due the depending on what happened half a mile before but at any point of time it could be the case that this is basically the at end at which we have them out of poverty but when we did Sitek searching the rent of of the condition or and few that exactly the same but
2:31:22
I'm not cost processes are homogeneous if the transition probability from some state to the other state will lose transition from believes he is the parent of of the time when it happens so I'm if you know the initial probability pie and no in what state your hand then just make the hop and this is kind of homogeneous just say if a point 6 all but 1 of it does not depend on come on what he did it before and then you can determine the overall distribution of the of the
2:32:07
complete processes you just go the the processes this the 1st random variable having some value and the last the red of the area with the last of the face having some certain that it is just given by the 1st observation being made so you are in the 1st 8 at some point this basically comes comes down to the bomb to the starting probability but for this kind of thing thinking across the want them to see so this is exactly the same Last that after he would in the 1st state He went from the 1st day to the 2nd stage which basically means that you have the state transition of care until you get into last 8 but seems as a mock of prose not depending on their that happened before but just 1 run step before which is basically 5 for the 1st than a year after the probability by 0 1 of in the next step and it goes on like that on to the pitch he said and then you make last step into the fine out state and this depends only on the Web you actually came from sell this is the last and lines on UK and this is the last transition Paul bility from the last day you and if you take a product of that the total probability of this instantiation of random area will happening with respect to your mock of model given by the initial probability distribution and all over the transition probe ability OCA yes with it but so
2:34:58
I'm in mock of more of time and so have the observation probability so it's not only interesting up the seconds of what states you go through a but also had a problem with that is that that in every state you make the observation so while I basically do is you take the homogeneous mock processes with the state's the and the transition culpability class the stops distribution in what states to stop at last a statistic prose that creates the observations that you do it with the offer of a sum of probabilities so being in state you made observations 0 up with the idea of this completely defined the mouth model the funeral every step probabilities defined you can't calculate everything and
2:36:17
the the weight of calculating it up we will do next time because it quite complicated and we need some nice Levite Algorithm for left out because of growth that the very many ways of putting together a different different observations probabilities was different transition probabilities I'm basing on all nostalgic for this is that again like a dynamic Programming very many ways to a 2 2 gets to what the high probability of this book next week excited you want to know how it works but we know we got the next solid next week Mexicans all them so well if you remember the holiday compete we you and but so this lecture we stop who was created by a humming and look at how we can represent a melodies and match melodies bypasses codes and with a little bit of fun every time walking and frame this supplementation were introduced in some ideas of democracy model from model you than some of its list of state machines basically that kind of like the producer of a 2nd of state over time and connect the 2nd all states to sequence of all of the shoots and next week we will
2:37:45
deal with Hull to compute the total probabilities of mob of model and step on to the tourist Prevot so out of the 3rd part of the match the video yes it the job of order and the yes it the and the walk yes you can to complete amenities to began off concept the mob of model of where the songs that sounds like a June for example so he of different states that reflect the different notes the different now like the different pitches of cost you can't of but questions get Celesio
2:38:40
in 2 weeks