Hello,

Is there a Linux tool that can retrieve sound-bites from an audio mp3 file given a sound pattern to search ? Consider over a hundred mp3 files of media announcements by one or more public figures where I want to 'grep' out particular spoken words eg any spoken instances of the words rice, drought and water etc. The analogous function is pdfgrep to pdf files as a way of finding pertinent documents.

Thanks,
Fabian

Hi, welcome here, at LQ

Interesting idea. I guess there is no software to do that "in one". But you may try to use a speech recognition software to convert mp3 to text and you can use grep on that.

Not quite what you want, but there's taggrepper.

I just put "audio grep" into a search engine, and the first hit is a tool called audiogrep which transcribes to text then searches that.

However, for matching audio snippets I would expect a better version to be something that directly compared audio waveforms (or signatures derived from them) - there's a bunch of online services which presumably use this method to provide music identification.

It didn't take much to identify the term "acoustic fingerprinting", and combining that with "linux" produced another result: EchoPrint, from a company which got bought by Spotify, but the code is still available at https://github.com/spotify/echoprint-server and https://github.com/spotify/echoprint-codegen - the latter appears to be a tool that "indexes" audio files.

There may well be other options out there - the above is only from quick searches that should have been the minimum performed before asking this question.

SongRec looks like a close match (warning for GitHub link though). It describes the overall process so depending on how comfortable you are with coding, it might be possible to build something similar.