FL ClearVoice

Audio enhancement nodes for ComfyUI featuring speech enhancement, super-resolution, denoising, and restoration using multiple AI backends.

Features

• Speech Enhancement - Remove noise and improve speech clarity
• Super-Resolution - Upscale low-quality audio to higher sample rates
• Audio Restoration - Fix degraded or compressed audio
• Multiple Backends - ClearVoice, Resemble-Enhance, and VoiceFixer models
• Automatic Resampling - Handles sample rate conversion automatically
• Mono/Stereo Support - Preserves original channel configuration
• Centralized Model Storage - All models download to ComfyUI/models/clear_voice/
• Real-time Download Progress - Progress bars with speed and ETA for model downloads

Nodes

Node	Description
Model Loader	Downloads and caches enhancement models
Process	Processes audio using the loaded model

Installation

ComfyUI Manager

Search for "FL ClearVoice" and install.

Manual

cd ComfyUI/custom_nodes
git clone https://github.com/filliptm/ComfyUI_FL-ClearVoice.git
cd ComfyUI_FL-ClearVoice
pip install -r requirements.txt

Quick Start

1. Add FL ClearVoice Model Loader and select a model
2. Connect to FL ClearVoice Process node
3. Provide input audio
4. Run the workflow

Models

ClearVoice Models

Model	Type	Output SR	Notes
MossFormer2_SE_48K	Enhancement	48kHz	High quality speech enhancement
FRCRN_SE_16K	Enhancement	16kHz	Fast speech enhancement
MossFormerGAN_SE_16K	Enhancement	16kHz	GAN-based enhancement
MossFormer2_SR_48K	Super-Resolution	48kHz	Audio upscaling

Resemble-Enhance Models

Model	Type	Output SR	Notes
Resemble_Enhance	Full Enhancement	44.1kHz	Denoise + enhance + vocoder
Resemble_Denoise	Denoise Only	44.1kHz	Fast denoising

VoiceFixer

Model	Type	Output SR	Notes
VoiceFixer	Restoration	44.1kHz	All-in-one audio restoration

Models download automatically on first use to ComfyUI/models/clear_voice/:

ComfyUI/models/clear_voice/
├── clearvoice/           # ClearVoice models
│   ├── MossFormer2_SE_48K/
│   ├── FRCRN_SE_16K/
│   └── ...
├── resemble_enhance/     # Resemble-Enhance models
└── voicefixer/           # VoiceFixer models

Requirements

• Python 3.10+
• 8GB RAM minimum (16GB+ recommended)
• NVIDIA GPU recommended (CPU and Mac supported)

Platform Notes

• NVIDIA GPU: Full acceleration for all models
• Mac (MPS): ClearVoice uses MPS, Resemble-Enhance falls back to CPU due to conv1d channel limits
• CPU: All models supported, slower processing

License

MIT