MelSpectrogram: Use given n_fft to initialize fb immediately #246
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
vincentqb
reviewed
Aug 19, 2019
| pad=self.pad, window_fn=window_fn, power=2, | ||
| normalized=False, wkwargs=wkwargs) | ||
| self.mel_scale = MelScale(self.n_mels, self.sample_rate, self.f_min, self.f_max) | ||
| self.mel_scale = MelScale(self.n_mels, self.sample_rate, self.f_min, self.f_max, self.n_fft // 2 + 1) |
There was a problem hiding this comment.
Is there a way to read off the size of the spectrogram instead of recomputing it?
There was a problem hiding this comment.
Not unless the spectrogram computes a specgram and we look at the tensor's dimension
There was a problem hiding this comment.
If that constant n_fft // 2 + 1 is used in a few places within Spectrogram, we could attach it to Spectrogram. If that's our inferred default dimension, we could also just attach a default_dimension to Spectrogram, in case that ever changes, but that would add clutter to the Spectrogram interface.
Alright, I'm ok with having the value recomputed here. If there's ever a change in this, the test will fail :)
vincentqb
approved these changes
Aug 19, 2019
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In the issue #245, it was determined that for MelSpectrogram the transform already knows how many bins the stft will have and thus can initialize the fb matrix without needing to infer the dimensions of its first input.
The benefit of initializing the fb matrix immediately as opposed to lazy load when given a first input is that loading and saving from state dict (
load_state_dict) will not throw an error as the tensor sizes match