Add Typecast SSFM-v1 (super realistic tts voice)

#39
by zironycho - opened

Hi guys.

Here is the super realistic TTS - Typecast SSFM-v1.
You can easily integrate this TTS API via Typecastai Python Package.
This is a proprietary model, but I would like to provide an API Token for TTS Arena
Please feel free to ping me for the API Token.

zironycho changed discussion title from Adding Typecast SSFM-v1 (super realistic tts voice) to Add Typecast SSFM-v1 (super realistic tts voice)
zironycho changed discussion status to closed
zironycho changed discussion status to open
TTS AGI org

Hi, are there any plans to open source this model?

@mrfakename Hi, there is no plan to open source the Typecast SSFM. In your project, there is only one proprietary model now. I think a better ranking would be achieved by comparing various models regardless of license.

From what I have heard, it is not better than the arena's current proprietary ElevenLabs model. It is only in the arena to be as the quality model that open source TTS strives to be. IMHO, no other proprietary models are needed for the arena. Unless it outclasses ElevenLabs. This does not.

@Pendrokar Your view of quality is probably right. But there are many different people on the planet with many different ideas and many different perspectives, and they may not all think the same as you and I do. Isn't that why we want to do this ranking? Isn't it better to compare how one model is better than another, rather than just comparing WER? I think quality is not just one dimension, but there are many dimensions. Some people think that the quality is good when the sampling rate of the speech is high, some people think that the quality is good when the speech is read as they think in their head, and some people think that the quality is good when the pronunciation is correct. Because there can be so many different dimensions, I think LLMs are also evaluated on different points when comparing them. Even the LMSYS Chatbot Arena Leaderboard, which you may find inspiring, has many different licenses.
I hope this project will grow to be as much of a challenge as LMSYS by comparing more different models from different angles.

TTS AGI org

Hi @zironycho - thanks for the request - we're currently slammed with requests and are at capacity with respect to the models on the Arena atm; we'll prioritise this request after the next release! Thanks again, and I am looking forward to SSFM-v1 being hosted on the Arena soon.

Sign up or log in to comment