search

eloimoliner / denoising-historical-recordings

A two-stage U-Net for high-fidelity denoising of historical recordings
MIT License
93 stars 12 forks source link

The music characteristics of the model #7

Open slliugit opened 1 year ago

slliugit commented 1 year ago

For this work, I have a problem hoping you can answer——The proposed method does not seem to be specialized for music denoising, it has no characteristics unique to music recordings. Thanks!

eloimoliner commented 1 year ago

Hi, The model was trained with the MusicNet dataset, which is mostly classical music. What kind of music recordings are you thinking about?

slliugit commented 1 year ago

My confusion is that this work does yield good results in the music denoising task, but looking at the model structure, I don't think it shows that it is specifically for denoising music. If this model is trained using the relevant dataset for speech denoising, could it also achieve a good denoising effect? Thanks!

eloimoliner commented 1 year ago

I think so. I did not apply any music-specific assumptions apart from the dataset, if this is what you mean. So if you train it with speech data it will probably work too.