Hello.What type of quantitative information makes up a song? I'm trying to analyse songs and create a list of numerical data (or signature) of the song. What information should I look? Thanks

It's an interesting question that I too would like to know the answer to. Services such as Shazam can identify a track over a mobile phone even when it's played in a noisy nightclub (yes, I have tried this) so it is obviously highly tolerant of noise in the signal.

I would be quite surprised if the system didn't involve both transient detection and a frequency transform, however.

I also just noticed you replied to your other post. The link you cite there is very interesting, but basically he was just taking the frequency at various points and feeding it into the neural network. That seems a reasonable approach.

Hi Kylotan,
I agree that site was interesting but I feel there could be more aspects than just the frequency. I was playing a game called Audio Surf and it was really interesting how the levels were developed. I was wondering if there are other aspects, or more over, how the frequency data could be used in order to extract even more information.

What sort of information could you want? Everything has to come back to frequency and amplitude in the end as that's the only data that is encoded. You can attempt to extract semantics by mining that data, eg. through recognising pitch or transients, and from that trying to guess at key, tempo, and time signature, but it's not clear whether that is what you want or not. They certainly don't uniquely identify a song.

Hi,
I managed to find out about the MFCCs of a song. Apparently, they're quite good representations of a song's signature. http://en.wikipedia.org/wiki/Mel-frequency_cepstral_coefficient
They won't necessarily identify the song, but it can give similarities between songs.
Thanks for the help