dpwe / audfprint

Landmark-based audio fingerprinting
MIT License
544 stars 122 forks source link

update to use gzip-compressed precompute features #55

Open ezavesky opened 5 years ago

ezavesky commented 5 years ago

Attempt to shrink precomputed feature files. To keep it simple, changing the default mode to include compressed as the desired format, but another command arg could also be added if preferred.

Interestingly, the peak-point only mode did not gain from the small collection of sample files. This is a quick table of files produced during the test run of the main Makefile.

 60K    gz/precompdir/Nine_Lives
140K    non-zip/precompdir/Nine_Lives
 52K    gz/precomppkdir/Nine_Lives
 52K    non-zip/precomppkdir/Nine_Lives

On longer files (9 files with average length of 1h15m) the savings are more apparent (using 11k, density 100, 1 shift).

 10M    gz/precompdir/audio/2d
 15M    non-zip/precomp/audio/2d
3.1M    gz/precompk/audio/2d
4.1M    non-zip/precompk/audio/2d