I was working with a singer for years but he died recently. I have hours of his singing, could I create a voice for him?
There is no support for custom voice databases.
I am in the same situation (my singer is still here with us, but he has completely lost the use of his voice) and I have tried to contact Dreamtonics via [email protected] , as @claire says…, but it seems almost impossible to get a response…
For the creation of a voicebank, specific scripts are used to make sure that there’s full coverage of al the phonemes.
Each script is repeated at various pitches, to ensure good coverage of the vocal range. The recordings need to be clean (free of background noise) so that the neural network doesn’t “learn” to add that noise as part of the sound.
The singer needs to be set up the same distance from the microphone, singing with the same dynamic level and style, with the same recording equipment setup.
So even with hours of training data, it’s unlikely to be in a format that Dreamtonics needs create a good voicebank.
In the future, voice cloning of singing voices will be possible. But we’re not yet at that point for commercial products.
You may want to try some open source projects such as DiffSinger, nnsvs, etc. But these are based on some AI models for research purposes but not stable or efficient enough for production environment.
Actually if there are enough quality singing data(like hours), especially, in this case, a singer’s dry voice, you don’t need more data to train an AI voicebank, so that’s not a problem.
“We”? But this is not a response from Dreamtonics, right?
I think the recording requirements are somewhat looser for AI development than Standard. The issues you mention were cited as a reason Eleanor Forte was delayed for so long; Volor was unable to get the voice provider to Japan for recording due to travel restrictions.
They were only able to proceed with development once they decided to make an AI voice database instead, because it could be recorded remotely without supervision. (source: Eleanor Forte (Synthesizer V Studio) | SynthV Wiki | Fandom)
There is certainly the fact that recordings made in different environments or with different equipment might compromise the consistency and quality of the voice, but if it’s being made as a private thing instead of a commercial product that could be a non-issue.
“Private voicebanks” have been made for Vocaloid in the past, but I expect an obstacle for doing similar with SynthV would be that currently Dreamtonics has to be part of the development process for every voice database, and they likely do not have the staff to dedicate to projects that will not be made commercially available.
I’m starting to wonder if there should be some sort of FAQ because these types of questions keep getting repeated ^^;
The main difficulty there is that we lack moderation capability without the support of Dreamtonics, and Discourse locks posts after a month so they can’t be maintained.
I think Discourse normally allows people with the “Regular” role and above to create “wiki”-style posts for this sort of thing, but as far as I can tell that’s not enabled for these forums.
edit: Wait i may have found the wiki button… I’ll try to figure it out later today or tomorrow.
Best of luck to you!