The Lyrebird Project

laststand6522732

As you may know, Lyrebird is a project to develop an open source version of Adobe Voco.  I've been following it with interest, as it could be an answer to the issue of voiceover talent.  A new feature has appeared on their website.  You can now generate a model of your own voice, and use it to read text you have entered.  I've attached an example.

At this point we cannot edit the inflection.  I am expecting that to change.  I heard an artifact in the recording too.  Still, I think it's promising.  Maybe you could try it and post a link here.

https://www.youtube.com/watch?v=suxGPY4RUBI

laststand6522732

OK.  You're wondering, "What good is that to me?  I only have access to my owh voice simulation, but my animation has three characters.  How can I collaborate with other voice authors?"

There is a way.  I'll bet you can guess what it is, too.  Anybody want to try it?

agent unawares

Yeesh, mine sounds really young. I'm thirty. https://lyrebird.ai/g/gs5ZOevq

laststand6522732

agent unawares said:

Yeesh, mine sounds really young. I'm thirty. https://lyrebird.ai/g/gs5ZOevq

Most of the models I see on this site are 30 or less. 

https://www.daz3d.com/gallery/#images/276956/

Yes, the artifacts are there.  Personally, I still like it better that any TTS I've heard.  It does sometimes struggle with sentence endings, and it is happiest when you really articulate.  I noticed they have a limit of 300 stored training recordings, even though only 30 are required for a basic test.  Someday I will make more than 30 and see what happens.

 

agent unawares

Rottenham said:

Yes, the artifacts are there.  Personally, I still like it better that any TTS I've heard.

It is very good in my opinion. Hopefully inflection support comes soon.