❓ Custom voice for German

[ChrisDelClea]

❓ Questions and Help

Hi @snakers4,

great package! Is it possible to add a custom voice for German and if so, how?

Best regards
Chris

[snakers4]

Hi,

Now a voice requires about 10-15-20 hours of recorded speech (see this for example).

We will be working on reducing this in the next iteration, but I doubt that the minimal amount will be less than 3-5 hours.

If you are interested in recording your voice / custom voice in German - I guess you can just use Thorsten as a reference.

If you will choose to publish this voice and / or share it privately with us - we can add a new voice to the project.

[ChrisDelClea]

Hi @snakers4 ,

thanks for fast response.
I think I could arrange 10h of recorded speech even though the background might be a bit noisy.
3-5 ofc would be far easier! It's for an internal project in my company so I am not sure but can ask.
Do you have a first entry point of how Thorsten did it?

Best regards
Chris

[snakers4]

though the background might be a bit noisy.

Typically TTS requires no noise in the background, otherwise the quality degrades.
Of course you can use denoising, but usually it is just easier to invest a bit more in the recording (US$100 microphone).
Our speakers recorded in the studio (Ksenia) and at home (Aidar and Baya) and I cannot really say that the studio voice is better, though people recording at home did it in a quiet room with microphone with a pop-filter.

Do you have a first entry point of how Thorsten did it?

I guess you can ask him directly @thorstenMueller (he is also active in his own repo mentioned above).
Also note that his dataset contains no stress, which ideally improves quality even further.

It's for an internal project in my company

Typically we discuss commercial inquiries in dm, please reach out to [email protected] or to @snakers41 in telegram