Any idea how i can go about implementing an algorithm for identifying a song, by matching it with a clip of the song or human voice or tune. I dont expect anyone to write out the algo here :), i know it involves a lot of things together, but i would like to know how i can get started. What are the technologies that i will need to consider for doing something like it. I am looking at something similar to Shazaam or Midomi. Also, is there an online service that does the same and provides API for doing it.
Asked
Active
Viewed 4,546 times
5

royhowie
- 11,075
- 14
- 50
- 67

PavanMysore
- 189
- 3
- 12
-
2The term acoustic fingerprint is often used to describe this class of problems, using that search term should help you locate relevant literature. – fvu Jun 11 '12 at 15:32
2 Answers
6
I saw a presentation at the devoxx conference about this, have a look at it here. And here is a blog post, by the speaker, about the same subject.

pgras
- 12,614
- 4
- 38
- 46
-
I didn't plan on being distracted for an hour, but that was a very interesting presentation... now.. where was I ... – Louie Sep 24 '14 at 10:05
1
Oh, it simple.
Split song into pieces, run FFT on each, extract few main magnitudes and store them as a hash with time information.
Then do the same with recorded audio and match with stored data with regards to time.
Simple isn't it? To be honest it's more complicated but idea is similar.

pawelzieba
- 16,082
- 3
- 46
- 72