Versatile Audio Resolution – improve audio quality. A service to bring a low quality compressed file to rich deep sound. The video shows the difference between “before” and “after”, this neural network really pulls it off.
It works with both voice and music recording – it can really help when working with “hissing” voice recordings. The service is on GitHub, and there are instructions for launching it there as well. (https://github.com/haoheliu/versatile_audio_super_resolution)
Description & Functionality
AudioSR is a comprehensive tool designed to upscale any audio input to a high-resolution 48kHz output. This essentially means it can transform your old, grainy recordings or low-quality clips into high-fidelity sounds that are pleasing to the ear.
What’s even more impressive is its versatility. Whether you’re looking to enhance music, speech, or ambient sounds like rain or even a dog barking, AudioSR has got you covered. The project’s objective is clear from its tagline: “Pass your audio in, AudioSR will make it high fidelity!” And it isn’t limited to a specific sampling rate, which makes it a truly versatile tool for audio aficionados.
Features
- Scalability: Works seamlessly across different types of audio, from speech to ambient sounds.
- Compatibility with Various Sampling Rates: No limitations based on the original audio’s sampling rate.
- User-Friendly Interface: Offers both batch processing and single audio file processing.
- Customizable Output: Users can select between different model checkpoints, choose computation devices, and even adjust parameters like DDIM sampling steps and guidance scale.
- Community Support: An active community on discord provides a platform for users to share their samples, thoughts, or raise any issues they might encounter.
How to Use
Getting started with AudioSR is fairly straightforward. First things first, you need to set up your environment. Optionally, you can create a virtual environment using conda:
conda create -n audiosr python=3.9
conda activate audiosr
Next, install the necessary package:
pip3 install audiosr==0.0.6
After the setup, you can start using AudioSR to enhance your audios. If you have a batch of files, simply use:
audiosr -il batch.lst
The results will be saved in an ./output directory by default. For single audio files:
audiosr -i example/music.wav
For a detailed list of all the command-line arguments and customization options, audiosr -h provides a comprehensive guide.
Future Endeavors
The developers are also looking to continually improve the tool, with plans to integrate a gradio demo and optimize the inference speed.
Acknowledgment
This incredible project is the brainchild of Haohe Liu, Ke Chen, Qiao Tian, Wenwu Wang, and Mark D Plumbley. Their dedication to the field is evident in their detailed research paper titled “AudioSR: Versatile Audio Super-resolution at Scale” which has been made available as a preprint on arXiv.
Conclusion
AudioSR provides an exceptional tool for those looking to breathe new life into their old or low-quality audio clips. Its versatility, combined with an active community and consistent updates, makes it a must-have tool for audio enthusiasts and professionals alike.
