This set of C++ programs are used to evaluate various audio identification algorithms to identify the radio station. Amongst others, it will be used to evaluate Shazam fingerprinting, Cross Correlation, and Relative Matching. It is concerned with various metrics such as how accurately each algorithm identifies the station, how much noise it can withstand, how quickly it can identify, etc. This set of programs can be used to answer questions about the cloud-based identification, local locking and local scanning algorithms.

For each of these functions, we will use the same wrapper code which evaluates some metrics on each of them. Also, for each system, the algorithm either takes a full recording and crops out the portion based on time syncing.

Parameters

• Fan out size
• Window size of each time sample
• Some threshold for detecting peak in spectrogram
• This takes as input the ground truth file and samples

Operation

• Takes the spectrogram of the audio sample
• Finds prominent peaks
• Then looks for other peaks in the fan-out region
• Takes the peak location and offset to create hash
• Matches based on alignment of fingerprints in sample and ground truth

Variants

• Just based on count
• Fully based on alignment

• Top secret.

• libsndfile
• Alglib
• STL
• Boost
• uBLAS