VALL-E 1.0: Reproduce the results of the paper, Train larger models, and greatly improve the speed of token extraction.

[lifeiteng]

VALL-E 1.0 Release notes

• Reproducted Demo https://lifeiteng.github.io/valle/index.html
  • Traning commands egs/libritts#demo
  • Inference valle/run.sh#L68
• Support Two stages training update modules #59
• Support Mandarin Mandarin AISHELL1 #85 Thanks @zhaomingwork
• Speedup Encodec AudioTokenizer AudioTokenExtractor batch inference #81 Batch Encodec #88 Thanks @Guokr233
• Thanks reviewer @zjwang21

---

This discussion was created from the release VALL-E 1.0: Reproduce the results of the paper, Train larger models, and greatly improve the speed of token extraction..