Seeking help in making algorithmic changes to an existing open-source python 2.7. component dejavu -- which is generating an audio-fingerprint for MP3 files/data-snippets. The audio-fingerprint algorithm is e.g. used by shazam to identify unknown audio/music recordings.
In our scenario we have access to 10 seconds MP3 content of a music file (always the beginning) and we want to create a DB of audio-fingerprints from these 10 seconds in order to identify music for which we don't have ID3 data and only an un-descriptive filename.
The (dejavu) audio-fingerprinting algorithm is able to detect music even with a relative high signal to noise ratio, but this is not the scenario that we are dealing with -- and therefore we don't need the large redundancy from the generated data coming from that approach.
We are seeking a freelancer who is creative enough to imagine a very similar algorithm that is better adapted to our scenario and that is also applicable to video identification (i.e. from extracted audio)
The actual extraction of MP3 from audio/video files is already done. We only need an algorithm that is extracting identifiable signals from 3,4 or 5 secs of MP3 audio in order to have a very high confidence level that we have identified the right music in a large DB of audio-fingerprints.
The product deliverable should be provided in Python 3.x.
The dejavu code is in [login to view URL] -- additional docu on audio-fingerprinting can be found in: [login to view URL]