Starstruk Logo MarkStarstruk

v0.1.1-alpha

usememos/memosv0.1.1-alphaFeb 23, 2024by yuantuo666

AI Summary

This alpha release introduces new vocoders, dataset preprocessors, and evaluation tools while fixing various bugs in training and inference workflows.

Key Highlights

  • Added DiffWave Vocoder implementation
  • Added HifiTTS data preprocessor
  • Added support for VALL-E pre-trained models (6k-hour Librilight)
  • Added Resemblyzer and WavLM for speaker similarity evaluation
  • Added Docker installation support

New Features

  • DiffWave Vocoder Added
  • HifiTTS data preprocessor added
  • VALL-E pre-trained model support
  • Resemblyzer and WavLM speaker similarity support
  • Docker installation support
  • Visualization support added

Full Release Notes

## What's Changed
* Update README.md by @eltociear in https://github.com/open-mmlab/Amphion/pull/41
* Add Demo Badges for SVC, TTA, and TTS by @RMSnow in https://github.com/open-mmlab/Amphion/pull/42
* Avoid Unbound Case of the download_root by @Adorable-Qin in https://github.com/open-mmlab/Amphion/pull/48
* Provide DEMO guide and express thanks by @Adorable-Qin in https://github.com/open-mmlab/Amphion/pull/55
* DiffWave Vocoder Added by @VocodexElysium in https://github.com/open-mmlab/Amphion/pull/56
* fix cosine_schedule_with_warmup for VALLE training by @HeCheng0625 in https://github.com/open-mmlab/Amphion/pull/52
* Added HifiTTS data preprocessor by @zyingt in https://github.com/open-mmlab/Amphion/pull/53
* fix a bug for vocoder inference by @VocodexElysium in https://github.com/open-mmlab/Amphion/pull/65
* Fix issues with while loop and trailing slash due to using sh instead of bash by @YasienDwieb in https://github.com/open-mmlab/Amphion/pull/60
* Fix Compatibility Issue with 'accelerate' Package by Reverting to Version 0.24.1 by @HarryHe11 in https://github.com/open-mmlab/Amphion/pull/73
* Add Resemblyzer for Speaker Similarity Evaluation & Bug fixes by @Merakist in https://github.com/open-mmlab/Amphion/pull/75
* Custom dataset & resume training recipe for SVC task by @viewfinder-annn in https://github.com/open-mmlab/Amphion/pull/72
* Fix bug for issue 76 (Import VariableSampler error) by @HeCheng0625 in https://github.com/open-mmlab/Amphion/pull/82
* Metrify RawNet3/Resemblyzer as Keywords & Update READMEs by @Merakist in https://github.com/open-mmlab/Amphion/pull/85
* remove redundant codes and update the function for fs2 feature by @ChenX17 in https://github.com/open-mmlab/Amphion/pull/86
* Adding Contribution Guideline for Amphion by @HarryHe11 in https://github.com/open-mmlab/Amphion/pull/92
* Check & Update PR Template by @HarryHe11 in https://github.com/open-mmlab/Amphion/pull/96
* Add issue templates by @yuantuo666 in https://github.com/open-mmlab/Amphion/pull/98
* Add WavLM speaker similarity for evaluation by @HeCheng0625 in https://github.com/open-mmlab/Amphion/pull/97
* Add AudioCaps dataset link for TTA by @HeCheng0625 in https://github.com/open-mmlab/Amphion/pull/100
* Delete utils/whisper.py by @HarryHe11 in https://github.com/open-mmlab/Amphion/pull/102
* Accelerate the calculation for CER metrics by @wsywsywsywsywsy979 in https://github.com/open-mmlab/Amphion/pull/104
* Fix bug for VITS resuming training by @lmxue in https://github.com/open-mmlab/Amphion/pull/108
* Add VALL-E pre-trained model trained on 6k-hour Librilight by @lmxue in https://github.com/open-mmlab/Amphion/pull/101
* Add preprocessing scripts for the librilight datasets by @HarryHe11 in https://github.com/open-mmlab/Amphion/pull/107
* Implement VitsSVC resume training / finetune feature by @viewfinder-annn in https://github.com/open-mmlab/Amphion/pull/95
* MFA Restructure & Environment Bug Fixes by @Merakist in https://github.com/open-mmlab/Amphion/pull/121
* Update VALL-E prompt examples by @lmxue in https://github.com/open-mmlab/Amphion/pull/126
* Update DiffComoSVC by @Lokshaw-Chau in https://github.com/open-mmlab/Amphion/pull/135
* Refine the multilingual front-end processing module by @lmxue in https://github.com/open-mmlab/Amphion/pull/137
* fix: G2P module fails to initialize #138 by @yuantuo666 in https://github.com/open-mmlab/Amphion/pull/139
* feat: support Docker installation by @yuantuo666 in https://github.com/open-mmlab/Amphion/pull/140
* Add support of visualization by @lmxue in https://github.com/open-mmlab/Amphion/pull/141
* Multi-speaker VITS & Hi-Fi TTS dataset structure by @zyingt in https://github.com/open-mmlab/Amphion/pull/131

## New Contributors
* @eltociear made their first contribution in https://github.com/open-mmlab/Amphion/pull/41
* @zyingt made their first contribution in https://github.com/open-mmlab/Amphion/pull/53
* @YasienDwieb made their first contribution in https://github.com/open-mmlab/Amphion/pull/60
* @HarryHe11 made their first contribution in https://github.com/open-mmlab/Amphion/pull/73
* @Merakist made their first contribution in https://github.com/open-mmlab/Amphion/pull/75
* @yuantuo666 made their first contribution in https://github.com/open-mmlab/Amphion/pull/98
* @wsywsywsywsywsy979 made their first contribution in https://github.com/open-mmlab/Amphion/pull/104
* @Lokshaw-Chau made their first contribution in https://github.com/open-mmlab/Amphion/pull/135

**Full Changelog**: https://github.com/open-mmlab/Amphion/compare/v0.1.0...v0.1.1-alpha