Spaces:
Running
on
T4
Apply for community grant: Academic project (gpu)
Hello,
We've released 🍵 Matcha-TTS which is a fast and efficient TTS architecture trained using recent advancements in probabilistic generative AI. Backed by diffusion and continuous normalising flows it is a fast and efficient architecture. It is relatively faster than other TTS architecture even on a CPU. However, its inference speed can be greatly increased using a GPU making it much faster than real time. If provided a GPU grant for the demo it will help others realise how fast and efficient a simple TTS architecture backed by strong mathematical formulations and engineering can be. It doesn't require a huge GPU just something around 3-4GB and a few milliseconds to synthesise long waveforms.
Done! ✅
Oh wow, that was fast! Thank you :D
Is there a way to increase the space sleep time? A restarting space can take quite some time.
Dope! Setting it to 10 hours for now, we might reduce it later based on the overall views the space gets.