ClearerVoice-Studio
The open speech-processing studio: pretrained FRCRN denoisers, MossFormer separators, super-resolution, and audio-visual target-speaker extraction in one pip package.
About
ClearerVoice-Studio is Alibaba's attempt to do for speech processing what SpeechBrain did for speech research generally — one toolkit where the pretrained models just work. The package (pip install clearvoice) bundles state-of-the-art models across four tasks: speech enhancement (FRCRN and MossFormerGAN denoisers at 16 kHz, MossFormer2 at 48 kHz), speech separation (MossFormer2), speech super-resolution (MossFormer2 upsampling to 48 kHz), and audio-visual target-speaker extraction. Checkpoints download automatically from Hugging Face or ModelScope, and the repo reports millions of model uses on ModelScope (3M+ for the FRCRN denoiser as of late 2024). It also ships training scripts and SpeechScore, an evaluation toolkit, plus a paper (arXiv 2506.19398) documenting benchmarks and community impact — 4.5k GitHub stars and counting. The whole thing is Apache-2.0. If you want one dependency that covers denoise, separate, dereverb, upsample, and extract-a-speaker, this is currently the most complete open option.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Paper (arXiv 2506.19398, author-reported): FRCRN/MossFormer denoisers competitive on standard SE benchmarks; 3M+ reported ModelScope uses for the FRCRN denoiser
- No independent benchmark replication found
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (TIER verified, metrics-v1). Evidence: Tongyi Lab paper (arXiv 2506.19398) documenting the suite; public Apache-2.0 repo with 4.5k stars; models auto-download from HF/ModelScope with millions of reported uses. Strengths: the most complete open speech-processing suite in one package — enhancement, separation, super-resolution, AV target-speaker extraction — with genuine adoption. Docked for: (1) benchmark claims are author-reported with no independent replication found; (2) it is a toolkit of models, not one model, so per-model documentation is thinner than a dedicated repo. License: Apache-2.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Tongyi Lab paper arXiv 2506.19398; pip install clearvoice; FRCRN + MossFormer + MossFormer2 models auto-download from Hugging Face/ModelScope
- · Repo modelscope/ClearerVoice-Studio created 2024-11-12, ~4.5k stars; FRCRN denoiser reported 3M+ ModelScope uses
- · Covers enhancement, separation, super-resolution, and audio-visual target-speaker extraction in one package