search

magronp / bregmisi

Phase recovery with the Bregman divergence for audio source separation
6 stars 1 forks source link

Is this useful in a general minimum phase trasform of audio signals? #1

Open 7sharp9 opened 7 months ago

7sharp9 commented 7 months ago

Hi, I wondered if this was useful in the context of general minimum phase transforms of audio sinals for magnitude preservation ratgher than being used during seperation of sources?

Thanks!

magronp commented 7 months ago

Hi, Can you specify a bit what you mean by "general minimum phase transforms"? The goal here is really to perform phase recovery on top of each magnitude spectrogram for source separation.

7sharp9 commented 7 months ago

By general I meant methods like the hilbert transform or the real cepstrum methods of transforming an audio impulse to a minimum phase version while retaining the magnitude information.

magronp commented 7 months ago

I think it depends on how you formulate your problem exactly. This method is designed to solve a phase retrieval problem under a mixing constraint - check Eq. (1) in our paper. If you can formulate a similar problem, though with another transform, then perhaps an extension is possible. However, note that if you're interested on single-signal retrieval (no mixing constraint as done here), then a better starting place would be this repo and the corresponding paper. It's dedicated to this setup, and goes much more in depth in the optimization problem and algorithms.