Skip to content

Performance

Jump to bottom Edit New page
devonbryant edited this page Mar 30, 2014 · 2 revisions

The primary focus of Safe is performance and scalability. The following benchmarks compare Safe's performance with three other popular audio feature extraction libraries: Yaafe, openSMILE, and Sonic Annotator.

All of these benchmarks were run on a Macbook Pro, OS X 10.7.5 with a 2.5 GHz Intel Core i7 processor and 8 GB 1333 MHz DDR3 RAM. For a more detailed description of these tests and results, see Scalable Audio Feature Extraction.

These benchmarks measured the total time taken to extract the following feature set:

  • Mel-Frequency Cepstal Coefficients
  • Spectral Flux
  • Spectral Shape (Centroid, Spread, Skewness, and Kurtosis)

The full dataset consisted of 1,332 songs totaling 5,120 minutes of audio, encoded in CD-quality 44.1 KHz 16-bit stereo WAV format. The following bar plot shows the overall time taken by each library. Safe-1 is a single-threaded configuration and Safe is the default (multi-threaded) configuration.

The following plot shows the time taken by each library for increasing numbers of songs:

This last plot shows the overall time taken by each library for single songs (1 - 10 minutes in length):

Add a custom footer
Add a custom sidebar
Clone this wiki locally