[YurtleTheTurtle]

I thought most companies were moving towards NN for this. Doesn't Shazam rely on a ton of audio processing and not NNs? Seems like a backward move.

[colordrops]

Their processing works perfectly - why would you need an NN?

[Godel_unicode]

I don't know if you know, but NNs are better than anything at everything. /s

[Jach]

NN would be more useful maybe for "recognize this song by me humming it, da-na-na'ing its main riffs, or badly singing a few words".

[colordrops]

Yeah that's true. I would hope they would make that a different function though. Google uses NNs for the android keyboard autocorrect now and it will incorrectly change correctly spelled words that I wrote 3 words back because it decided I wanted to write a more "likely" sentence. It replaced "worry about that" with "sorry about that". The shotgun approach of NNs in certain scenarios is infuriatingly frustrating. Sometimes an exact match is preferable to a probabilistic method.

[skndr]

The data set could be really useful.

[nerfhammer]

Apple has rather good music data, too.

I'm assuming they just want one more icon that leads to Apple Music and not to Spotify

[abritinthebay]

NN isn’t a magic bullet. You still need data, inputs, and users.

Branding is nice too.