feat!: when preprocessing, everyvoice forces equal length time and f… #421
Conversation
This file contains 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
roedoejet
force-pushed
the
dev.ap/fix-audio-and-spec
branch
from
f756ce9
to
145b98e
Compare
|
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #421 +/- ##
==========================================
+ Coverage 73.45% 73.49% +0.03%
==========================================
Files 43 43
Lines 2837 2852 +15
Branches 467 468 +1
==========================================
+ Hits 2084 2096 +12
- Misses 668 671 +3
Partials 85 85 ☔ View full report in Codecov by Sentry. |
roedoejet
changed the title
…ature reps audio must be divisible by the declared hop size the number of frames in the spectrogram must exactly equal the number of samples in the audio when multiplied by the hop size
roedoejet
force-pushed
the
dev.ap/fix-audio-and-spec
branch
from
145b98e
to
a96307d
Compare
SamuelLarkin
approved these changes
May 9, 2024
| @@ -741,6 +759,14 @@ def process_text( | |||
| return (character_tokens, phone_tokens, pfs) | |||
|
|
|||
| def process_spec(self, item): | |||
There was a problem hiding this comment.
We should annotate the return type
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
…eature reps
audio must be divisible by the declared hop size
the number of frames in the spectrogram must exactly equal the number of samples in the audio when multiplied by the hop size
PR Goal?
We had very tiny mismatches between length of the audio and Mel spectrograms when the number of samples in the audio was not evenly divisible by the hop size. This PR truncates the audio when preprocessing to be evenly divisible by the hop size.
This is implemented for both the input and (potentially) upsampled audio in the case of vocoder that applies super-resolution to input spectrograms.
This means that we effectively discard up to
hop_size - 1samples in the audio. For the default (22.05 kHz audio with 256 hop size) this means that we discard up to 255/22050 = 11.5ms of audio, which I think is OK :)Fixes?
Hopefully makes it easier to some downstream models that depend on exact alignment between
Feedback sought?
Sanity checking
Priority?
low/medium
Tests added?
Added some unittests
How to test?
Just having a look at the code is sufficient. If you want, you could try preprocessing some data and ensure that the resulting audio is divisible by the hop size (256 by default). Similarly you should be able to read in the preprocessed spectral features and multiply the 2nd dimension by the hop size to obtain the exact number of samples in the audio
Confidence?
high
Version change?
If we weren't in pre-alpha yes.