Audio Super Resolution

Another pending project of mine was ASR. Audio Super Resolution is a dream for a person like me, who wishes to listen to old mp3 files in much higher fidelity than possible. While ASR is more like DLSS / ISR for image, I think it is good to have options even if they may not be the best.

I tried a lot of projects for this and kept on failing. To my surprise this project worked with decent results and pretty straightforward steps. My only issue was that it requires 16GB VRAM, so instead of doing this on GPU, I had to do it on the CPU (my primary 16 GB GPU is busy with OLLAMA at the moment, and this can be done decently fast on a CPU also).

The Project I am talking about is Versatile Audio Super Resolution.

audiosr -i example/music.wav -d cpu #only change I had to do to force CPU because of the VRAM issue

P.S. quantization does not work well here, since audio is much more sensitive to distortion. I still hear some distortions in the generated samples, but it is better than other options.