hi-paris

hi-paris / wavlm-vocoder-french

Public

WavLM-to-Audio neural vocoder for French speech reconstruction — layer ablation study and adversarial supervision as a foundation for continuous voice conversion (JEP 2026)

hi-paris.github.iowavlm2audio-demo french-speech tts voice-conversion wavlm-vocoder
19
0
100% credibility
Found Mar 30, 2026 at 19 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

This repository implements a neural vocoder for reconstructing high-quality French speech waveforms from WavLM representations, supporting training, evaluation, and inference as part of academic research on voice conversion.

How It Works

1
🔍 Discover the French Voice Rebuilder

You stumble upon this fun project while exploring speech tools, with a live demo showing amazing French audio recreations from voice patterns.

2
💻 Set It Up on Your Computer

Follow the easy guide to download and prepare everything, so your computer is ready to play with French voices in minutes.

3
🎵 Feed It Your French Audio Clips

Gather some French speech recordings, like stories or conversations, and let the tool learn their unique voice patterns.

4
⚙️ Train the Voice Magic

Hit start to teach it how to rebuild clear, natural-sounding French speech from hidden voice features – it runs smoothly on your setup.

5
🔊 Create New Speech Samples

Use your trained tool on new audio to generate fresh French voice recreations that sound incredibly real.

🎉 Enjoy Lifelike French Voices

Listen to the high-quality results, perfect for experiments, demos, or sharing your cool voice conversion discoveries with friends.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 19 to 19 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is wavlm-vocoder-french?

This Python project builds a neural vocoder that reconstructs high-quality French speech waveforms from frozen WavLM representations. It tackles decoding self-supervised speech features back to audio, serving as a foundation for continuous voice conversion pipelines. Users get CLI tools to train, infer, evaluate metrics like PESQ/STOI/MCD, and run layer ablation studies with adversarial supervision.

Why is it gaining traction?

Its JEP 2026 acceptance brings academic rigor, with ablation studies showing optimal layer selection (e.g., last 9 layers) and GAN training yielding 15-25% metric gains over baselines. Developers dig the ready configs for no-GAN baselines vs. full adversarial setups, plus HF pretrained models and a live demo for quick French speech reconstruction tests. Multi-GPU support and chunked inference handle real datasets efficiently.

Who should use this?

Speech ML researchers experimenting with WavLM for French TTS/VC, voice conversion engineers needing a reconstructive decoder before latent-space manipulation, or French ASR devs validating feature quality via waveform regen. Ideal for those with PyTorch setups training on corpora like Common Voice or M-AILABS.

Verdict

Solid research starter with excellent docs, tests, and MIT license, but 19 stars and 1.0% credibility score signal early alpha maturity—expect tweaks for production. Grab it for French speech experiments if you're okay forking for scale.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.