MeanVC2 Quality-Only Checkpoints for OpenVox

This repository is a reorganization of third-party checkpoints for use with OpenVox. It does not contain a new or modified model.

It includes only:

MeanVC2 120 ms + 40 ms quality checkpoint
160 ms Fast-U2++ ASR encoder
Required Vocos checkpoint
Required WavLM checkpoints
Corresponding MeanVC2 configuration

The original Python inference code and its dependencies are not included.

Included Files

File	Purpose
`ckpts/pretrained_models/meanvc2_120ms_40ms.safetensors`	MeanVC2 120 ms + 40 ms voice-conversion checkpoint
`preprocess/ckpts/fastu2pp_160ms.pt`	Fast-U2++ 160 ms ASR encoder
`ckpts/vocos/vocos.pt`	Vocos vocoder checkpoint
`src/config/config_120ms_40ms.json`	MeanVC2 120 ms + 40 ms configuration
`preprocess/ckpts/wavlm_large.pt`	WavLM Large checkpoint
`preprocess/ckpts/wavlm_large_finetune.pth`	Fine-tuned WavLM checkpoint
`preprocess/ckpts/wavlm_large_cfg.pt`	WavLM configuration extracted from the verified checkpoint

Provenance

MeanVC2, Fast-U2++, Vocos, and MeanVC2 configuration: https://huggingface.co/ASLP-lab/MeanVC2
WavLM weights: https://huggingface.co/k2-fsa/TTS_eval_models
WavLM configuration: Extracted locally from the verified wavlm_large.pt checkpoint.
Original MeanVC2 initialization/setup: https://github.com/ASLP-lab/MeanVC2/blob/main/initialization.py

Licensing and Attribution

This repository contains files originating from multiple upstream sources.

No single blanket license is asserted for all files in this repository. Individual files remain subject to the applicable licenses, terms, and redistribution requirements of their respective upstream projects and sources.

Before redistributing or using these files, review the licensing and redistribution terms of each upstream source.

Preserve all applicable upstream:

Copyright notices
License notices
Attribution
Usage restrictions
Redistribution requirements

The organization of these checkpoints for OpenVox does not change the ownership or licensing of the original files.

SHA-256 Checksums

File	SHA-256
`ckpts/pretrained_models/meanvc2_120ms_40ms.safetensors`	`04ea21a4fb55400469e6004ae842c3e7f059f759e92946fb95123418cad5d818`
`preprocess/ckpts/fastu2pp_160ms.pt`	`9a372c1b7178f564a15da1cd36901d7d0c9081a2fabf53cb6cc0d971deca4928`
`ckpts/vocos/vocos.pt`	`df1f2ba9f7ac35c96832579421ee1ed913a68c3a1d2f8a6536739f902f194a93`
`src/config/config_120ms_40ms.json`	`1645ae2e51ee82eb1d53f8401e3a8ca9af79a4bd3ca6189b001491af53f531be`
`preprocess/ckpts/wavlm_large.pt`	`6fb4b3c3e6aa567f0a997b30855859cb81528ee8078802af439f7b2da0bf100f`
`preprocess/ckpts/wavlm_large_finetune.pth`	`51f07e3b94d9e0262a6a675ef5a087be3dd09e8c62e9d886827f44f82fe7f94b`
`preprocess/ckpts/wavlm_large_cfg.pt`	`d8c312a42f3a73e969894b59ae918312811d5e6f34744c0e970455af6fefcd38`

Notes

This repository is intended to provide the specific checkpoint layout required by OpenVox.

For the original MeanVC2 implementation, documentation, training details, and inference code, refer to the upstream MeanVC2 project.

Downloads last month: -; Downloads are not tracked for this model. How to track