Add diagonal_loading optional to rtf_power #2369
Conversation
|
@nateanl has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
@nateanl has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
No, it was not in the previous release, I think we can regard the changes as BC. what do you think? |
torchaudio/functional/functional.py
Outdated
| (Default: ``True``) | ||
| diag_eps (float, optional): The coefficient multiplied to the identity matrix for diagonal loading | ||
| (Default: ``1e-7``) | ||
| eps (float, optional): a value to avoid the correlation matrix is all-zero (Default: ``1e-8``) |
There was a problem hiding this comment.
can this docstring be more helpful? does a value added to the denominator in the beamforming weight computation. from #2368 make sense here, and is it worth adding that this is only used for the case when diagonal_loading=True?
There was a problem hiding this comment.
Makes sense. I will align the eps docstring in the functions and modules.
There was a problem hiding this comment.
The eps here is for diagonal loading, which is confusing with eps in computing beamforming weight. I decided to exclude it from the API and use the default value in _tik_reg.
|
@nateanl has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
When computing the MVDR beamforming weights using the power iteration method, the PSD matrix of noise can be applied with diagonal loading to improve the robustness. This is also applicable to computing the RTF matrix (See https://github.com/espnet/espnet/blob/master/espnet2/enh/layers/beamformer.py#L614 as an example). This also aligns with current
torchaudio.transforms.MVDRmodule to keep the consistency.This PR adds the
diagonal_loadingargument withTrueas default value totorchaudio.functional.rtf_power.