AudioSR Models for ComfyUI

Pre-trained AudioSR (Versatile Audio Super Resolution) models for use with ComfyUI-AudioSR custom node.

Models

audiosr_basic_fp32.safetensors

• Purpose: General audio super-resolution
• Best for: Music, sound effects, podcasts, mixed content
• Format: FP32 SafeTensors
• Size: ~5.9 GB

audiosr_speech_fp32.safetensors

• Purpose: Speech/voice optimized super-resolution
• Best for: Voice recordings, vocals, speech content
• Format: FP32 SafeTensors
• Size: ~5.9 GB

Usage

Installation

1. Install ComfyUI-AudioSR via ComfyUI Manager
2. Download model(s) from this repository
3. Place in ComfyUI/models/AudioSR/

Quick Start

ComfyUI Workflow:
Load Audio → AudioSR → Preview/Save Audio

Recommended Settings:

• Steps: 50-100
• Guidance Scale: 3.5-5.0
• Model: Use audiosr_speech_fp32.safetensors for voice, audiosr_basic_fp32.safetensors for everything else

What it does

AudioSR upscales low-quality audio to high-quality 48kHz output using latent diffusion. It:

• Resamples to 48kHz
• Enhances high frequencies
• Reduces compression artifacts
• Adds clarity and detail

Model Info

Based on AudioSR: Versatile Audio Super-Resolution by Haohe Liu et al.

Original repository: https://github.com/haoheliu/versatile_audio_super_resolution

License: MIT

Hardware Requirements

• GPU: NVIDIA RTX 3060 or higher (6GB+ VRAM minimum)
• RAM: 12GB+ recommended
• Works best with audio > 8kHz input sample rate

Credits

• Research: Haohe Liu et al.
• Paper: AudioSR on arXiv
• ComfyUI Integration: ComfyUI-AudioSR