| Technology Search |
|---|
| Your Cart | ||
|---|---|---|
|
| Customers | ||||||||
|---|---|---|---|---|---|---|---|---|
|
||||||||
Home 
Browse for Technologies Software Robust Landmark-Based Audio Fingerprinting - Academic
Browse for Technologies Software Robust Landmark-Based Audio Fingerprinting - Academic
 |
Robust Landmark-Based Audio Fingerprinting - Academic |
| Price: FREE |
|
These routines implement a landmark-based audio fingerprinting system that is very well suited to identifying small, noisy excerpts from a large number of items. It is based on the ideas used in the Shazam music matching service, which can identify seemingly any commercial music tracks from short snippets recorded via cellphones even in very noisy conditions. The basic operation of this scheme is that each audio track is analyzed to find prominent onsets concentrated in frequency, since these onsets are most likely to be preserved in noise and distortion. These onsets are formed into pairs, parameterized by the frequencies of the peaks and the time inbetween them. These values are quantized to give a relatively large number of distinct landmark hashes (about 1 million in my implementation). Parameters are tuned to give around 20-50 landmarks per second. |
|