Spaces:
Running
on
A10G
Apply for community grant: Academic project (gpu)
Hi Dear HF Team! 😃
This is an open-source implementation of Microsoft's latest Text-to-speech model VALL-E X 🎙️, from paper Speak Foreign Languages with Your Own Voice: Cross-Lingual Neural Codec Language Modeling 🌐. It is basically a 24 layers, 1024 d GPT-style model 💻. I have made a demo page about this model for your inspection and consideration 📋.
It takes about 60s to synthesize a 6s speech on free CPU ⏳, but only 2~3s on a single RTX 3060 ⚡. I sincerely hope that you could grant GPU resources for the Hugging Face space of this project, so that more people can have the chance to play with this awesome model 🚀.
Best Regards!🤗
Hello, great author:
I encountered a problem when I used the downloaded pre-training model to generate directly according to the basic usage method in the explanatory text: I generated blank noise audio. The same is true for audio generated by running python -X utf8 launch-ui.py into the user interface. However, the results generated by opening the online demonstration link you provided are normal, and I don't understand what went wrong. Vallex_checkpoint.pt and vocos models have been downloaded.