At Thu, 19 Apr 2012 15:46:34 +0200,
alex wrote:
> I was looking for an audio search engine when I realized that I did
> not know any that was as much known as google for text and document
> or vimeo for video or again flicker for still images. I read a short
> page on wikipedia that talks about 2 algorithm approaches, one
> searching in the metadata for keywords (not that interesting to me)
> and some others (without links) trying to differentiate speech form
> music in the body-content of an audio file for example. I thought
> the good place to ask such a question was here. Maybe you are all
> aware of projects like this, Apologise for not having read the
> archives.
The Online Music Recognition and Search project conducted at
Goldsmiths' College and Queen Mary, University of London delivered a
tool called AudioDB. It's a essentially a feature vector database
which uses locality sensitive hashing to find approxiamte nearest
neighbours to a given query vector.
http://www.omras2.org/audioDB
The code is very much research-grade; it works, but you have to
compile it yourself and there's little documentation. It's written in
C++ within bindings for a few other languages.
Another important caveat is that it's not (like Google, Flickr, etc.)
a search engine over an existing database. If you want to use it, you
have to create your own database. And to do this, you need a feature
extractor to get features from your audio collection. You also need an
audio collection. The same research project also produced a simple
extractor called fftExtract:
http://omras2.doc.gold.ac.uk/software/fftextract/
As an alternative, you may be interested in looking at the echonest
API, see especially their track API:
http://developer.echonest.com/docs/v4/track.html
It also occured to me to check if Shazam publish an API. I couldn't
find one.
Best,
Richard
Received on Thu Apr 19 2012 - 14:32:11 BST