Skip to content

Resamplers#176

Draft
GaganNarula wants to merge 1 commit into
mainfrom
gagan/resampling-methods
Draft

Resamplers#176
GaganNarula wants to merge 1 commit into
mainfrom
gagan/resampling-methods

Conversation

@GaganNarula
Copy link
Copy Markdown
Collaborator

Adding torchaudio resampler. #122
It is definitely an order of magnitude faster..

@GaganNarula GaganNarula requested a review from mil-ad November 18, 2025 09:32
@mil-ad
Copy link
Copy Markdown
Collaborator

mil-ad commented Nov 18, 2025

It is definitely an order of magnitude faster..

On GPU or CPU?

Also why are we doing this now? Is it a priority?

@nkundiushuti
Copy link
Copy Markdown
Contributor

sorry for intervening...since we noted difference in the results given the resampling method, this should be done cautiously, benchmarking with torch and the usual librosa method we used in the past to see if we get similar results

@mil-ad
Copy link
Copy Markdown
Collaborator

mil-ad commented Nov 18, 2025

yeah this changes the default to torchaudio so we have to careful

@GaganNarula
Copy link
Copy Markdown
Collaborator Author

yup you guys are right! thats why its only a draft PR. However, we should discuss this soon because for most of the datasets in esp-data on-the-fly resampling is required because a) there aren't already existing sample rate versions in the buckets (like iNaturalist) and b) resampling with librosa is about 300x slower (not a full benchmark, just taking one file 90 sec file and resampling it 100 times).

@GaganNarula
Copy link
Copy Markdown
Collaborator Author

It is definitely an order of magnitude faster..

On GPU or CPU?

Also why are we doing this now? Is it a priority?

cpu test

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants