I understand the idea, thank you. 

I picture the interaction with the UI as follows: 

- The user enter notes and lyrics as they normally would, only by setting
the Instrument to "Virtual Singer". Additional settings would depend on the
synthesis tool.
- Each staff inside a virtual singer that has lyrics is separated to make an
independent input for the singing synthesizer. 
- Synthesis step. (to be discussed later) 
- The output audio need to be separated into notes probably by time
calculation (tempo, note durations, ...) I am currently looking more in
depth into the fluid component to see how to achieve this. (I would be very
grateful for developers' inputs/advices on this)
- The score is played starting from the cursor (like MuseScore does in other
situations). 
- Repeat the steps after each modification. 

As for the synthesis step, I am left with two (maybe three) options. To
summarize the previous discussion, they are: 

- Using Sinsy (open-source) :

Pros: 
- Direct input.
Cons: 
- Works only for Japanese. Adding support for other languages might be
possible, but the lack of compatible data could be hindering; 
- Lacks quality compared to the two following options.

- Using Sinsy ( web-service)

Pros: 
- Direct input;
- Supports Japanese, Chinese, and more importantly English (even though the
accent isn't all that good, I don't think this is a big problem);
- The output's quality is more than decent. 
Cons: 
- No guarantee of unlimited availability;
- Requires an internet connection;
- The web service in itself isn't all that slow, but depending on both the
user's internet connection and the file's size, the delay could get really
big (waiting for a response, then downloading the audio (compressed, in the
number of staves, then uncompressing it), then processing it...);
- No possibility of adding other languages for us (unless the developing
team does).

- Using v.Connect-STAND: 

Pros: 
- A lot of possibilities (in voices, languages, accents, ...), only limited
by the big Utau voice database on the internet.  
- Could give very good results if used correctly (and with the proper voice
set)
- Japanese works directly, and it is possible to add support for several
other languages (English, French, ...) .
(- Maybe a possible performance optimization is possible.)
Cons:
- Using it requires to hack into it to solve several usability issues.
- Indirect input (would require something to generate its meta text
sequences from scores and the good settings, with the amount of work this
could require)
- There's hardly any doc, and code comments are all in Japanese (while I can
read it with some effort, it is clearly not ideal)
- Adding support for any language, especially English, will make us face
problems about phonetics and converting to the used voice set's own lyrics
format. 
- A lot of voicebank are Virtual singing enthusiasts' product, with unclear
licences and random quality. Looking for something adequate (for MuseScore
and a precise language) could be a very tiring matter (and so would
producing our own set of voice). 

Other interesting tools I can think of right now: 
- eSpeak could come in handy to add support for European languages (since it
can convert pretty smoothly from text to phonemes), whatever the
synthesising tool we choose to use.
- kakasi can do the Romaji/Kana conversion, so we could allow both inputs
for Japanese.
- iconv, convmv, and everything else that makes dealing with file encoding
easier (for some tools).

The ideal solution of course would be to combine all the aforementioned
tools and then add some more, and I honestly do want to. But the time
constraint makes this completely unrealistic, so a good first step would be
to make the smallest functional thing, and since anything functional
requires English, Sinsy's web service may be the most obvious first step.
How acceptable is the delay, though? We should probably impose a waiting
time limit (and maybe consequently a file size limit). Developers' opinion
on this would be great! 


benjisan wrote
> I'm a choir master and a voice teacher, and i'm agree with that idea : we
> are looking for the easiest solution!
> I actually do as many of other voice teachers and choir masters : using
> Musescore for editing sheets, and  using Harmony Assistant rival for audio
> examples. But we all would like to use Musescore without Harmony
> Assistant!

Could I please also have your opinion on the suggested interacting scenario?
Is there something you could use and see missing there? 

As the deadline is dangerously closing in, I will try to make a draft by
today, I will take into account any additional suggestion. 

Thank you very much. 



--
View this message in context: 
http://dev-list.musescore.org/GSOC-2016-Regarding-the-Virtual-Singer-project-idea-tp7579698p7579753.html
Sent from the MuseScore Developer mailing list archive at Nabble.com.

------------------------------------------------------------------------------
Transform Data into Opportunity.
Accelerate data analysis in your applications with
Intel Data Analytics Acceleration Library.
Click to learn more.
http://pubads.g.doubleclick.net/gampad/clk?id=278785351&iu=/4140
_______________________________________________
Mscore-developer mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/mscore-developer

Reply via email to