Hi!
I have done something similar in the past using fingerprinting to prevent duplication inside Amazon S3 buckets. Going on this experience I believe it can be extended to suit your use case.
A few questions, however:
1) Once we have retrieved the sonic fingerprint of the file, do you plan to use a service to look up the artist and title or are you going to trust the details from the stream?
2) The more we can record from a stream the better the fingerprinting will be and the higher the accuracy. This will rely on either a timer interval or waiting for the metadata to change. Due to piracy metadata sometimes changes late which may not get us the best fingerprint.
Please do not hesitate to contact here on Freelancer should I be of further assistance to you.
Warmest Regards,
Stephen Lombard