7 Comments
User's avatar
Andrew Smith's avatar

Nice, man - very clear.

Do you know much about Soundhound, and if it operated on similar principles? I remember using this app in like 2008 and it worked, sort of.

Expand full comment
Gold Bassey Edem's avatar

Ahh I've never heard of Soundhoudn man. I didn't have access to a mobile phone by 2008 man, haha.

But the use of acoustic fingerprints are the standard for most audio recognition algorithms.

Expand full comment
Andrew Smith's avatar

I love that old history, how we got to some of these modern devices and programs. Thanks for giving me things to think about!

Expand full comment
Zan Tafakari's avatar

Great post! Crazy how long spectrogram matching has been around though - wouldn't be hard to surpass it with a bit of deep learning sprinkled on top?

Expand full comment
Gold Bassey Edem's avatar

Glad you liked it man. How long have they been around bro? I recently just came to know about them.

Ahh I don't see how deep learning could work here. Perhaps a classifier?

Expand full comment
Zan Tafakari's avatar

They've been around for a very long time haha - I saw some people on twitter talking about how they could build something better but i can't find the tweet any more...

Yeah I think a classifier would work pretty well - maybe even on sub-components of the signal

Expand full comment
Gold Bassey Edem's avatar

Yeah, Yeah. A classifier could work good. But you'd still need to get the audio fingerprint as the input feature for the model.

I guess the question is would a model inference be faster than a database query in this case.

I think Spectograms could probably be replaced with some sort of system.

Expand full comment