Open source TTS model: Mars 5

Speech and Generative AI Text To Speech (TTS / NLG)
1

I recently discovered this open-source TTS model Mars 5 and invited the team from camb.ai to talk more about it in the Mozilla AI discord.

Wanted to see what folks here think about it. GitHub - Camb-ai/MARS5-TTS: MARS5 speech model (TTS) from CAMB.AI

4 Likes
2

Very cool demo! The online demo link in their readme appears to be dead, but I’m curious to see what kind of hardware this requires and how fast it is. I’ll also note that the AGPL license can limit integration with projects that use less restrictive licenses (i.e. BSD, Apache 2.0).

As I understand from my brief reading through the repository, this would enable support for a variety of different voices without training a full model for each one? This would be interesting for customization, in particular I’m thinking about this voice/WW changing feature that @mikejgray is working on.

1 Like
3

Hey @NeonDaniel, thanks for the feedback.

:re LICENSE, as we mention in the README, we are open to taking requests for using it under a different license. Folks simply need to email us. We’ve received a few hundred emails and are responding to each one.

Apologies for the online demo being down, we’re quickly moving towards a HF-hosted demo which should be much more reliable.

1 Like
4

Thanks for the quick response and flexibility with licensing; I missed that when I skimmed the repo yesterday.

I have only good things to say about HuggingFace so far for demo hosting; we have a few demos running there and they have been very reliable :slight_smile: .

1 Like