Deep learning tool audioFlux: a systematic audio feature extraction library | Hacker News Reader

Deep learning tool audioFlux: a systematic audio feature extraction library

github.com

62 pointsCMLab8 comments

A library for audio and music analysis and feature extraction, which supports dozens of time-frequency analysis and transformation methods, as well as hundreds of corresponding time-domain and frequency-domain feature combinations, can be provided to the deep learning network for training, and can be used to study the classification, separation, music information retrieval (MIR), ASR and other tasks in the audio field.

I spent a good part of high-school trying to make my house lights into a giant VU meter. I feel almost obligated to re-do those old projects again, using this library.

Thanks for releasing this!!

Interest is the best teacher. I hope it will be better and better in the future.

Such a lovely statement.

I had a bit of a philosophical naval gazing about hoping we get a computing that was more apparently interesting to everyone again, that was engageable. Just over a silly-but-fun Parallax Mountains background generator[1]. I hope you don't mind if I share it, because that carrying a torch for interesting ideas, for letting people learn & see & get hooked & excited is such a great fuel for people.

As another great post, "What You Get Is The World"[2] was on, we need rich material to grow & thrive & gain interest. Tech keeps having repetetive interactions, where the "user" just pushes buttons. It's not really computing, it's an interaction-appliance. I hope we can create a computing where people are allowed to be interested again, where we can see how stuff works, go in & monkey around. I believe that's what it takes to let interests form, to let us keep walking the path to better & better futures.

Thanks for your words, I agree very much. I hope we can feed the roots, make accessible the world, so we can stay interested in it, all be allowed to keep bettering that future. You've done well helping that along, cheers, thanks again.

[1] https://news.ycombinator.com/item?id=34916424

[2] https://theconvivialsociety.substack.com/p/what-you-get-is-t... https://news.ycombinator.com/item?id=34161077

Thanks for sharing! I’ve been using some TTS to make audiobooks for my kids. Sometimes there will be weird artifacts (sounds like someone brushing against a microphone), or the voice will change from masculine to feminine during quotes, or it will struggle to speak a phrase and make guttural sounds.

My next task is to figure out where this might be happening so I can rerun those segments. The books can be hours long so it’s hard to catch.

Could audioFlux be used to support identifying some or all of these?

According to your description, it should be the problem of spectrum generation.

AudioFlux itself does not directly provide TTS function, but it can be used to analyze spectrum problems, or try different spectrum types. The effect of using erb/bar type spectrum is better than mel spectrum.

I don’t need help making the TTS, but I need to know when it’s a weird generation.

Based on your description this may be what I need but I clearly have a lot to learn

Very nice. Thank you.

Reply on news.ycombinator.com