Free Software Virtual Singer

Video in TIB AV-Portal: Free Software Virtual Singer

Formal Metadata

Free Software Virtual Singer
Title of Series
CC Attribution 3.0 Unported:
You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor.
Release Date

Content Metadata

Subject Area
Virtual Singers such as Hatsune Miku are very popular in Japan. Before that there was also an MBROLA-based Virtual Singer called Melissa and a multilingual formant synthesizer called Virtual Singer by the French company Myriad. All those Virtual Singers are based on proprietary software, until recently there was no free replacement. In 2011 I discovered that the Music Technology Group had released their Spectral Modeling Synthesis library under the GNU GPL, and I decided to use that one in my own UTAU compatible resampler.
NP-hard Maize Computer animation Order (biology) Speech synthesis Cuboid Endliche Modelltheorie Musical ensemble Local ring Hypothesis
Complex (psychology) Computer program Parsing Length Java applet Latin square State of matter Multiplication sign Sheaf (mathematics) Numbering scheme Set (mathematics) Solid geometry Mereology Food energy Formal language Subset Impulse response Preprocessor Optical disc drive Estimator Component-based software engineering Envelope (mathematics) Atomic number Hypermedia Different (Kate Ryan album) Cloning Software framework Damping Office suite Endliche Modelltheorie Extension (kinesiology) Position operator Stability theory Physical system God Area Covering space Boss Corporation Arm File format Structural load Software developer Closed set Sampling (statistics) Electronic mailing list Maxima and minima Lattice (order) Value-added network Message passing Process (computing) Phase transition Order (biology) Duality (mathematics) output Row (database) Point (geometry) Interpolation Observational study Variety (linguistics) Code Spektrum <Mathematik> Power (physics) Revision control Frequency Goodness of fit Population density Causality Internetworking Well-formed formula Metropolitan area network Form (programming) Boolean algebra Time zone Standard deviation Gender Uniqueness quantification Total S.A. Sphere Film editing Computer animation Logic Personal digital assistant Speech synthesis Musical ensemble Object (grammar) Window Local ring
Computer program Beat (acoustics) Wage labour Computer file Multiplication sign 1 (number) Water vapor Drop (liquid) Mereology Metadata Formal language Twitter Element (mathematics) Goodness of fit Mathematics Energy level Software testing Contrast (vision) Form (programming) Area Standard deviation Shift operator Structural load Consistency Physical law Content (media) Sampling (statistics) Planning Volume (thermodynamics) Leak Computer animation Software Integrated development environment Personal digital assistant Speech synthesis Local ring Row (database)
Dialect Local area network
Point (geometry) Presentation of a group Shift operator Multiplication sign Equaliser (mathematics) Sampling (statistics) Time series Mereology Demoscene Formal language Population density Logic Personal digital assistant Hypermedia Internetworking Speech synthesis Normal (geometry) Right angle Software testing Musical ensemble Window Form (programming) Row (database) Social class
Slide rule Computer program Presentation of a group Context awareness Open source Computer file Multiplication sign Formal language Goodness of fit Population density Well-formed formula Videoconferencing Endliche Modelltheorie Address space Social class Information Software developer Projective plane Moment (mathematics) Content (media) Compiler Personal digital assistant Website Musical ensemble Resultant Library (computing)
Computer animation
I mean it will be a I did my master's thesis about speech and singing wasn't mn this year false corn 2016 I have been present some unlocked the I didn't there than currently going on thinking isn't it as the so the 1st 1 its had inspired me is there's often evoke a lot this r of art are hard Mamiko C is actually uh choosing of then she is very popular in Japan and there is the best book on local model using incentives itself that local or at most developed by the amount and by the University pompous have some music style technology called then local as solid in box better provides created in a 200 NATO always not easy go get detuner me goes off and vocal orders only licensed 2 companies does partnership voices and Thomas are companies did not get the license
so there's still no John Vocaloid if and for French and development zigs and uses technology based on hidden Markov models for was chosen and months you'll has the the license for after you need tools that can stop the coding and usually the old quite voices from existing singers was often form the known I mean man voice actors and then after the recording the voice data needs all be processed using complex spectral orders and debt is done by local knowledge in the background then then you want to understand How did the spectral models slack then you can lead various papers the published by yeah marked the music technology goal and on along as a master Japanese unique because it is and vocal load also has a classical later talk them on can input melody and low legs and while singing the as as including m on our next section it it along the the things the the you for in the model the the you you but end the flat vocal-noise them was at all can call or not that is also a momma's speech and desires out it is not designed for same but it can be used as as they now but someone a tried easy access and then annual clear voice databases this is not so much easier f you don't need to manually Mark voice parsers as voices periodic signal I can demonstrate as laid down at the end in the plants of the nope has mark marketing is needed but and this impulse to voice quality of bows genders and Warner and local Lord a concatenation based it they conquered amended the segments that have been preprocessed I move nominated it there's a spectacle preprocessing but in the case of and 1 of the just clinical up there the Odyssey His older no phase mismatches can obj your and in In both cases don't put Morris sounds similar to the input voice and note that formation is applied so the Vocaloid cool boss on the similar to adapt and then you all has simple English and born novelist as a festival voice and Bold voices sound similar their time a and balsam desires as are actually compared with dual citizen resides in the time of on and on can be done on and all at internal continues free of X and local loaded similar atoms the trial more modern them such as the quot total and then what I will some glitches unpublished uh z and still be around a subset of 2 Vocaloid DSP so when I look on a local AT-clone I the also cleared and and what clone that's can be used by other communities such as stable and accessibility the this is a screenshot off the cut into often which is yeah free and localized for the Japanese language and the on the BM and allows you to point all data would tell the components using an value but there's also the connect standard that is designed to all the plays part of on tell and this works only using that energy and the although it's of van local clone still club cited as often as their anyone Ghana court voices that voice-band formatted as documented and that's please after which can handle those was banks the then formalized processing what uses the exemplars that's beware centrally postings things 1 of those things is changing the pitch and the answer is the engine and the length of speech segments the and and there are several others leaving is the samplers which are based on but which was at that time released under the GPL and 1 of those that toys although and is more than an leaves and is the gonads stand there is also a tool called laughter on edge conquered amounts to those of of a size that Waldo clear that is controlled by the all it all by a cut in G then you want all cleared voices you can use them as often as all them all and that prompt those are least as these often about officer only on the run on Windows in there's another back that makes those programs only but on unknown in Japanese systems then there are some vocal installed on demand non-Japanese systems mostly due the fact that they use the pop attended Japanese researchers and coding this is only used on and also the later I discovered a poker and quality it kind of makes that is the singing isn't is formed at it was all nearly that Lehtinen pile using media of all pose and must mass stored effort of the and there are some since and users examples on the Internet using that these big fondant economics on those endpoints some and Warner voices and leaves and leaders support for far In the old flies festivus speech and desires on and later I decided to tool detected a porch and the logic it kind of makes using violent it is now a kind of mixed 5 it kind of looks tool and there's huge out does that is my own laid it on using George at I want 1 version which used and as a out these and I don't I had little that that also and 11 no I'm working on and what also owns clone that you can see more the natural and more and allows you to use existing model Moses for and does not natively on cannoli Knox it does not take fire in the Java allows you drop sees the shop and the
dentist check transport there is not a nice feature that is also present in local node to but 11 that is similar to leave right up many all start pardon 1 poker and all of that and pull commands begin playback and that is some colonized so variety by maximally Molly z is vocoder-based speech and intimacy wounded used in order you would always and last then Daniel marginal as new where the lessons that sample was similar to the Japanese invented tandem said and they're they're known and would either lesson I have that cover sphere of those the lessons the first one is that fast if CEO estimations all speech and speech signals are out of the object and contained them Nocedal and then if you out frequencies are suddenly it seals all higher frequencies of ought a perfect geoscientists so it should be it TO and this is used as the China was still around a fundamental frequency then using this cheap trick I localisms area spectral envelope is estimated that it takes the form speech powers land along various and then only text of power spectral doesn't take and all account and this power spectrum as small as using the known fundamental frequency at the at that point and a finally happening or density is estimated deals in the gold till they function and this goes to the power mu does it is possible at all so desires speech and using and was you cannot upon but speech closing name a by adding that using a form and and in the new version of about there is this arms and is using linear offers and myself have a gene we demonstrate here your this these things about us of the of the here and do I have as you also variety then did also you parliament does a can be transformed but you can always all saying and complex those panel made us fall later you was down by begin yours began yours and the Stepstone I talked at idea old off the become extent undesired up and 1st need there we used a concept from told the most indefinite and caused by the pedals as he did this is actually done in their Cheap Trick adolescent but then in the 2nd pass the you also again then combustors ignorant as we need minimal phase spectra for others and the also used as cepstral model created dollars many months from and now I explain home speech data as conquest and conquered embedded in 1 example which I had tested this using an older version version of God in uncompressed there are to the launch but bath of flame model clones tending solid spec Tolman form containing the old apple pear the odds C. and there is a lot of Latin dances and the state so the Congress this by the things that mel-cepstral and take only the 1st study tool samples and we don't need babble positions all it is vanished as a FIL float and this makes good condos speech to adopt the much even smaller than so again and the they size and the ties are recorded in CD quality so there's lots doesn't look that then add no outlets In this case the peaks and other sentences evidence would lock on but is it 1 is designed for on high-quality things and is and now I'll explain invite V Australia called delay involved the FFT phases that the plus minus pi thank you interval vijf causes of some practical problems dead then the phases on the edges this becomes complementary and it needs to be and that is all of the before the the formula had the D 4 or C and the lesson and different adolescent contemplating almost yours but all and emergence of flight then that he was still mn interpolation also fundamental she's still get the position of deposit but in that time zones as this could not be estimated so monopolists false dropped in the this and assessed speech and but he starts on the 1st was puzzle that that is normally done locally maximal move in the 1st was segmented and then after losing his propagated data to the next pitch spelling out now next to local maximum fault at P 0 and Daniel created this 1 spectral comb which contains the face of the input voice and that is taken was visited already known spectral and list them you the former you ever listen the don't care this scheme tendency on their face the calculated the apparent isn't it fall each frequency bands and as is similar to order spectral models and is this smallest stem opted the music technology called used in early versions of local knowledge but this 1 is like the and allows much better component although under voiceless the the the In the dollar explain how that's 1 record voice this as you have a good voice Paul sensing framework n fall down there is a there is political and all of a mortgage comes to this a list of Japanese recording lists these are usually based on the Japanese yen organized forbid that can be easily recorded for as a stand singing eventually speech you'll need need 4 the phone set and floors of language it is better to overcome this for the phone that and In only sustained policy of meetings and and order for example
eLex ghosts desires always and therefore it does not produce value good thing eyes so i is decided to leave i'd all other almost log Molina was using Japanese and and time idealist and pull out all played as a stimulus the diary fond altered I can who placed this 1 let's speak even if I dozing inventory of they added pitch offers and his lies the voice needs stole B more well adjusted to end that was wasn't possible overseas speaker at a time I usually use a condom Michael form this big membrane and popular and after the recording monophone and I calculate the pitch shielding was and if everything is OK it is involved in the base it's not IID heftily court is simple form and that is assisted by those of the but it is also possible to you always the existing was as often festival speech and his assistant this is also known to be able to using and it was used in the church software thing income Jordan after the recording voices the hospital transcribed devices so that you can in yours then for things inducers and there are several in combat include they still legend and according still fast just using here's are gonna and alter nobleman T if your Japanese then then we have to marks edit and the beginning and the middle also designed for name and for for Japanese you'll need change telling after Volunteer delaying of complement is never changes by contrast in things and stood still part much more languages that is and will and local law to be yours the International for Medical updated some violent for Japanese local extend give gonna gets mapped to a December was submitted end the then we can and here are gonna in and then they've started supports Japanese and this could could be mapped to a Germany's was German was the labeling of was said I still be done manually using them considers undated or no Hidden Markov model-based off them that is free and drugs that reliably In singing isn't isn't really have of plea out of on each levels all when you have a car and thinking the volume must start on the beat and dead test found all inserted in the labeled and if you will concatenated Wallace then you will have to specify and all then he she which depends on the government that entire end then you need all mark the segments metadata are stable or transient president of vowel consonant and only during Hobbes also the vowel a leak that can be sustained and I'm just saying not as common can be compressed so all of you and mark those things that you want to congress the but in other cases don't know others can be all so ain't folded it is very complex and I don't have channel always home to make it more easier here I have an example all such and Japanese would always called that's song any the 1st 1 is still colorful name which columns convex can valine M. M. area and forwards and now it is similar the and In this case you can see that the dead end is voice standard k is unvoiced and unvoiced segments like a cannot be stretched so there's not need still be market and is silence on the beginning the silences needed by the police and love but it may still be is kept trend going sentences solid also said and the kind of all the market in the or do we need no I have decided to home the the place dollar sinks after founded by phone dead bird is severely height volunteers it puts users a better quality than plot internal pitch shift down I mean decided told like to compliment fast and then build it was the colonies all of i to complement and allows me to use pre-recorded voices and currently on one's own fucking on a drop in inland placement for them on a executable and I was all modified these big that I can all explored labelled data and then any life but of did didn't new evidence on the east because what so I can stretch some segments and produce long of marvelous and I I have packet those those programs on my mind about society there was a kind content edges instead I load In the last 2 years this is von off everything is a isn't screenshot of my dollars isn't all of which currently can only use the speaker but I am plan to add water so partner end then for of other languages sentences English and I has to change the labor that are formatted the is to simply find what I'll only some applied Japanese laws are languages that have lesser complements that's finish it doesn't work for English and German in millions of them except I also the example of the speech pronouncing the name of 1 of those smoker load characters and these vacant puts up all of dolls labors the speech and before the that had at most impossible to all those things and is is this is paper and now I demonstrated GMM worn the connect stand and as but not focus on things and his then this 1 is the license plates for skip the didn't the and
here I have some the examples including once Swedish some of them all and at most done some years before local knowledge the move up the environment through the skin and people bring their own and all the money in the hands of the same in and in this can name to that I had a lot of them so and I have the Anthony and the most can I was here end in the also changed doing these be controlled via but also then you have 1 example so discussing sound as natural and because users for months and years the knowledge the this 100 thousand elements and creates those conquest highest matches only done once all the new land again and then in much faster on the and then on files are created this and here the start and other sample comes along the launch truck play about the absence of any of these things but
now it we you knowledge they still is various compressed size which are used frozen visitors will end
Punsters and it finds that created it starts that that he'll have some empty measure last and when we see it LAN which he had OK which the regions 98 when we created when we mean the play we a lead lining leading light 8 money seen in there I'm an body seed earn hand hand a by seeing and it at the 0 all in all of the of the but so just can be involved in quality by using a breathe
record speech samples of there's the as the speaker has only a z 100 kilobytes speech was then which is the out money many languages and the no and present some very the dialog window on the city the the the the the the the the the the the the this and then I can but we the right now so you only have a plot of land they forms
and know and played a faster my that is not a problem nearly finished uh I finished that presentation knowledge only that so if you allow a scene part and growing and me a to an we the about days perhaps are you see on me of them and they're sitting in pile of so assess time-series yeah where are the community in the center myosins too fast 1 is a sample by and for them equal devil cannot collect which I showed fast then note processing it was applied in all the West plot to pitch shift this end but it's not designed for music ended the point of MFN von wanted to use the scene pomeranian time you'll me at my fuck room when you happily is seen on the internet we use them because they're so seemed not right so that the guys probably bird is going to be in the same my and it is unique class does not alone tall that convert the female voice into a male voice and that can be done this but Vermont but I 1st show in normal in the pitch-shifted this is a a a lot scene the on the right to had In this case there was no artifacts which are caused by facing this momentous T. P. S all and a then the can also spectral he's going to well it's and period of much the was and in Vocaloid Thompson's is known as a friend might have sworn in the pool the E. R. acini all monomials on the we incur Lerman name Walt and so I obtained last long using revised and speci fying effect or at all to the time but I also transposed defocusing phone octave down so ordered it can be found by a major voice on but we end up he i have an example that means that's all it is also based on the logic based on did I learn and on and on the on the rest of 17 and eventually the and they had the way in the test 1 of the charts a much better sound quality density of indigenous knowledge and importance in the size of it is only for speech and then you will include the world and then in the hands can name and in the case of them 1 of the useful part much smaller languages such as leaders virtually no Japanese songs for and of the and no Media
and Ireland give you into the possibility to ask thanks to the 0 for your ovaries intensive and low an in-depth presentation and we I think we have for to 3 questions so that any and questions formulas so please hold on so you can talk to the mike can you please all fall is to give you can as courts the and the I got the in the in the data are are used that OK any more than you more questions so that if you worked in the basically on alongside the as your working basically on Japanese the some of the moment of time on a small easily at most existing knowledge so far only Japanese but I want to know what much more Benton so what are the necessary steps to of same supporting usual on German as well as the as the language the is smaller that is needed to and so on I usually form more developed for us supporting your work or is it more technical shows that needs to be address this is that is the attendance and I come nearly the short but it didn't have time and tool that knowledge and so all the time improves disease OK and is there a open source community working on these issues do you have an assignment but the community is only for all and for me a content pool on efforts and had to all the light on existing community class it's a good community after voice calls I is they thought the developer community is also valid the big user and community and you must have the heart of was by most into alignment program and those who are not programmers often right so sometimes they as of the speed of but in other cases there no so as it can be very popular as 1 of the this is the national government to begin with him on that 1 is monthly and you don't know what in the density community in New model created the replacement for that and without out synthesizer so there is no they communities that that inaccessibility community for the of the of the value the compiler and I'm using the music exist x and library of my old language he was as In this file value a compiler I decided to leave used as a result I don't have to hide my own OK so thank you very much so that do you have a slight of of some contact information so people 0 who made of what's a video of project you also somebody has a kind of home that and and some patterns in human and has OK the final go the slide down I have modal back site is this fall then I can be
contacted so easy got my nickname Lindsay in on the local and all of its most and yeah I can hold this quantity kind of it's called the mowing but I right think that 1 of the digital content in been published yet in the must address the context but I think the answer the OK excellent again thanks again at the time of across much of the health thank