README.md

metadata

title: Miipher 2 HuBERT HiFi GAN V0.1
emoji: ⚡
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 5.38.0
app_file: app.py
pinned: false
license: apache-2.0
python_version: 3.12.10
models:
  - Atotti/miipher-2-HuBERT-HiFi-GAN-v0.1

🎤 Miipher-2 Speech Enhancement Demo

This is a Gradio demo for Miipher-2, a high-quality speech enhancement model that combines HuBERT, Parallel Adapters, and HiFi-GAN vocoder.

Features

• Real-time speech enhancement - Remove noise, reverb, and other degradations
• Multilingual support - Built on mHuBERT-147 for 147 languages
• High-quality output - 22.05kHz audio output
• Easy to use - Simple drag-and-drop or microphone input

Model Details

• Paper: Miipher-2: High-Quality Speech Enhancement
• Model: Atotti/miipher-2-HuBERT-HiFi-GAN-v0.1
• GitHub: open-miipher-2

How to Use

1. Upload an audio file or record using microphone
2. Click "Enhance Audio" button
3. Download the enhanced result

Technical Details

The model uses:

• SSL Backbone: mHuBERT-147 (multilingual)
• Adapter: Parallel adapters inserted at layer 6
• Vocoder: HiFi-GAN trained on SSL features
• Input: Any sample rate (auto-resampled to 16kHz)
• Output: 22.05kHz enhanced audio

Citation

@article{miipher2024,
  title={Miipher-2: High-Quality Speech Enhancement via Self-Supervised Learning},
  author={Your Name and Others},
  journal={arXiv preprint arXiv:2505.04457},
  year={2024}
}