ComfyUI Extension: ComfyUI-AudioSR

ComfyUI-AudioSR Introduction

ComfyUI-AudioSR is an innovative extension designed to enhance the quality of audio files by upscaling them to a high-fidelity 48kHz output. This tool is particularly useful for AI artists and audio enthusiasts who wish to improve the clarity and richness of their audio content, whether it be music, speech, or sound effects. By leveraging state-of-the-art latent diffusion techniques, ComfyUI-AudioSR can transform low-quality audio into a more vibrant and detailed sound experience. This extension is seamlessly integrated into the ComfyUI environment, making it accessible and easy to use for those familiar with this platform.

How ComfyUI-AudioSR Works

At its core, ComfyUI-AudioSR uses a process called latent diffusion to enhance audio quality. Imagine your audio file as a painting with faded colors. Latent diffusion acts like a digital artist, carefully restoring and enhancing the colors to make the painting vibrant again. Similarly, this extension analyzes the audio file, identifies areas that lack detail, and fills in the gaps to produce a clearer and more detailed sound. It does this by resampling the audio to a higher frequency, enhancing high frequencies, and reducing artifacts that often plague low-quality audio files.

ComfyUI-AudioSR Features

• Audio Super Resolution: Upscales audio to 48kHz, enhancing high frequencies for a richer sound.
• Native ComfyUI Integration: Works smoothly with ComfyUI's Load, Preview, and Save Audio nodes.
• Spectrogram Visualization: Provides a visual comparison of audio before and after processing.
• Automatic Sample Rate Handling: Accepts various input sample rates and adjusts them to 48kHz.
• Stereo Support: Processes both mono and stereo audio, handling each channel independently.
• Long Audio Support: Uses smart chunking to process long audio files without length limitations.
• Model Caching: Keeps the model in memory for faster processing of subsequent audio files.
• torch.compile Optimization: Offers a speed boost for FP32 models through PyTorch compilation.
• VRAM Management: Option to unload the model to free up GPU memory between runs.
• Interruptible Processing: Allows you to cancel processing mid-run using ComfyUI's interrupt button.
• Progress Reporting: Displays a real-time progress bar to track chunk processing status.

ComfyUI-AudioSR Models

ComfyUI-AudioSR offers different models tailored for specific audio types:

• audiosr_basic_fp32.safetensors: A general-purpose model suitable for music, sound effects, and various audio types.
• audiosr_speech_fp32.safetensors: Optimized for voice and speech content, providing enhanced clarity for spoken words. Choosing the right model depends on the type of audio you are working with. For instance, if you are enhancing a podcast or a speech recording, the speech model would be more appropriate.

What's New with ComfyUI-AudioSR

Version 1.1.1

• Fixed tensor dimension mismatch for small chunks, ensuring smoother processing.

Version 1.1.0

• Introduced SageAttention support for faster processing on compatible GPUs.
• Added a dtype selector for better control over compute precision.

Version 1.0.6

• Resolved chunk positioning issues to eliminate volume drops in long audio files.
• Improved overlap-add normalization for consistent amplitude across chunks.

Troubleshooting ComfyUI-AudioSR

Common Issues and Solutions

• Model Not Found Error: Ensure models are downloaded from HuggingFace and placed in the correct directory.
• CUDA Out of Memory: Enable unload_model to free VRAM or reduce chunk_size.
• Poor Audio Quality: Adjust guidance_scale and ddim_steps for better results.
• No Output Audio: Verify connections in ComfyUI and check for error messages in the console.
• Slow Processing: Use torch.compile for a speed boost and ensure GPU usage.

Learn More about ComfyUI-AudioSR

For further learning and support, consider exploring the following resources:

• AudioSR Paper (arXiv) for an in-depth understanding of the underlying technology.
• Project Page for additional insights and updates.
• ComfyUI GitHub Repository for community support and discussions. These resources provide valuable information and community support to help you make the most of ComfyUI-AudioSR in your audio enhancement projects.