Surfboard: Audio Feature Extraction for Modern Machine Learning | Hacker News Reader

Surfboard: Audio Feature Extraction for Modern Machine Learning

arxiv.org

115 pointsjonbaer20 comments

I was curious, how does Surfboard compare to Librosa?

It looks like Surfboard is a wrapper combining feature extraction methods from Librosa and other libraries, plus some of its own implementations. Is that correct?

https://github.com/novoic/surfboard/blob/master/COMPONENTS.m...

Hi there, Author of the paper here. Thanks for your question. I have actually already addressed this question on our reddit post, please see it: https://www.reddit.com/r/MachineLearning/comments/gqvnpv/p_s...

Thanks, this helps!

> Surfboard actually builds on LibROSA for a few of its components (e.g. MFCCs) and its functionality (e.g. loading .wav files into memory). In that sense it is the same as LibROSA for a lot of their communal functionality.

> LibROSA was built to extract individual components for music analysis. The components we built into Surfboard are largely different to those in LibROSA (apart from a few). If you are interested, please read our paper and documentation to find out more!

Also pretty useful is Essentia [1]. It's created by the guys at Universitat Pompeu Fabra and designed for use both in research (for things like Music Information Retrieval) and in industry (iirc it's used in a guitar tuner app and a few other things). Written in c++ with python wrapper, pretty cool!

  [1] https://essentia.upf.edu/

This is interesting. Essentia has a js wrapper too that runs via WASM.

This is very off topic but why is something submitted May 2020 to preprint listed as https://arxiv.org/abs/2005.08848

The 2005. confused me - is it some coincidence of random number generation or something else I missed?

It's YYMM.ID I believe

Last year may was 1905.xxxxxx

Yep. This and the last year will be the only two confusing years for a century.

aha. Why are they trying to save 2 bytes worth of space? Is it legacy ?

Maybe somebody knowledgeable of the field can pitch in: How could this be used to assess accuracy of pronunciation in language learning contexts?

Github: https://github.com/novoic/surfboard

https://github.com/novoic/surfboard

I understand that the author(s) like to surf, but what does that have to do with this library?

Surfboards catch waves, like sound waves?

Sure. I wish they'd put a bit more work into the name. Maybe I'm thickheaded since I can't get over the word association each time I read, "Surfboard".

Could it be a reference to the SURF feature detector? I thought that was for images though

Perhaps they’ve been using the WaveSurfer software for audio analysis (https://en.m.wikipedia.org/wiki/WaveSurfer).

Or a reference to the AudioSurf game which used audio feature extraction to create levels?

Hi there, Author of the paper here! We chose Surfboard because 1) it's a fun name and 2) surfers ride waves, and Surfboard works with WAVEforms. That's it really :D

Reply on news.ycombinator.com