3 ms·
Related, what’s the current SOTA on STT models freely available? T5 is pretty good but the closed Google and Meta stuff seems better.
by benreesman 3y ago
Related, what’s the current SOTA on STT models freely available? T5 is pretty good but the closed Google and Meta stuff seems better.
- nielsinho 3y agoTorToiSe (https://github.com/neonbjb/tortoise-tts https://github.com/neonbjb/tortoise-tts) produces the best quality speech of any freely available model. However, its long inference times makes it impractical for voice chatbots like Gdansk.
- kjok 3y agoWhat's the reason for the high inference latency? Any ideas on how this could be improved?
- drewbitt 3y agoIt has been sped up, but still not fast enough for this use case. https://github.com/manmay-nakhashi/tortoise-tts-fastest https://github.com/manmay-nakhashi/tortoise-tts-fastest
- nielsinho 3y agoTorToiSe is composed of many large models: GPT-2 for text encodings, as well as a large VQVAE encoder + large diffusion model decoder. Only the big spaghetti inference code (+ weights) has been published, so there's a high entrance barrier for re-training / improving it.