Hacker News Reader

Comment by patrakov | Hacker News Reader

Does it rely on the fact that music has a concept of notes with well-defined frequencies and that there is no such thing as 216 Hz (must be 207.6 or 220 Hz) when the music is played on a properly tuned instrument? Or does it use other heuristics?

View on HN

I tried it out and the instructions have tips on what record to pick, they say to pick a well known version of a song (like not a live version etc), and preferably song with a beat, but it says it doesn’t use any 3rd party APIs or libraries, only Apple APIs. So my guess at to what it’s doing is using a ShazamKit recognition behind the scenes and looking at the frequencySkew value of the matched result. It also gives you one answer after listening, instead of a continuous gauge, which seems to corroborate song recognition. It probably won’t work with an obscure record that is not Shazamable. And so I don’t think it can measure wow & flutter as a result.

Still pretty cool for those that need to calibrate a turntable, or verify 33 vs 45 PRM for a record.

I love that everyone is guessing all these methods of detecting pitches or using the camera to count rotations, and it turns out that they're most likely literally just using a built-in API and displaying a return value [0].

0: https://developer.apple.com/documentation/shazamkit/shmatche...

I guess a phone's camera can be used to count the RPM, e.g. by filming the label and counting how many milliseconds it needs to do a full rotation (1818ms for a 33 RPM record)

I tried to build a prototype of something like this with opencv once, didn't get it to work reliably though. I have a feeling there should be a relatively simple signal processing version of this, basically a spatiotemporal Fourier transform, that should solve it.

I would go with something like sift features (or a non patented variant thereof if you plan commercial usage). You analyze a first picture and extract the center part. You run a feature detector on each image and the can cheaply match (scale and rotation invariant) against each other - this gives rotation and scaling.

Yea that's precisely what I ended up doing, but the frame extraction was quite finicky.

How can it use ShazamKit if it processes everything on-device?

I'm really really curious how this is done now...

They don't claim everything is done on-device, just that the audio stream processing is.

> Grooved does not collect any data, whatsoever. The audio stream is processed locally on your device and never recorded.

Which is consistent with how ShazamKit works [0]:

> Audio is not shared with Apple and audio signatures cannot be inverted, ensuring content remains secure and private.

0: https://developer.apple.com/shazamkit/

A "properly tuned instrument" still requires a base frequency (usually defined for A4). Whether to use 440Hz or another slightly different frequency is a fun debate to get into.

Older recordings on analog media also can have timing error at the recording stage or duplication stages.

Reply on news.ycombinator.com