3 ms·
I've actually been really curious, does anyone know what kind of algorithm they use to detect similarities? From what I've read of the tech behind Shazam/Sound
by crazygringo 8y ago
I've actually been really curious, does anyone know what kind of algorithm they use to detect similarities?
From what I've read of the tech behind Shazam/SoundHound, I was under the impression that the kind of audio fingerprints produced and indexed relied on the waveform/FFT and rhythmic input being identical, and that a band doing a really accurate cover would never trigger a match because it would be totally different in a million tiny details, the details used for the actual fingerprinting.
But clearly that's not the case here. No two pianists play perfectly alike, so what kind of algorithm is robust enough to match the underlying composition as opposed to the execution of it?
- Footkerchief 8y agoThere's got to be some fudge factor in algorithms used by e.g. Shazam -- an FFT of a microphone recording is never going to be strictly identical to the FFT of the source material.
- yomly 8y agoShazam predates the recent big Deep learning movement iirc. I can conjecture but I feel tech for "understanding" the music has come a long way since Shazam first started. Esp when you see YouTube labelling music in personal recordings. EDIT: also see [0] for a semi high profile case of a busker getting banned from FB for doing a cover [0]http://www.bbc.co.uk/newsbeat/article/39338488/busker-thanks-ed-sheeran-for-apologising-for-facebook-ban http://www.bbc.co.uk/newsbeat/article/39338488/busker-thanks...
- cormullion 8y agoLast year I received a YouTube copyright claim against my use of an extract from the Open Goldberg Variations: https://www.opengoldbergvariations.org https://www.opengoldbergvariations.org which I presume was generated by an automated process that couldn't accurately distinguish between performances.