Using executable extractors out-of-box


Although Essentia serves as a library, it also includes a number of examples that can be found in the src/examples folder apart from its main code. Some of them are executables that can be used compute sets of MIR descriptors given an audio file. When compiled they will be found in the folder build/src/examples and they are installed to /usr/local/bin (Linux/OSX).

The prebuilt static binaries for these extractors are available via Essentia website and can be used without installing Essentia library.


These examples include several executable command-line feature extractors that you might use to familiarize with the type of descriptors Essentia is able to compute or use them as a reference when building your own extractors:

  • essentia_streaming_extractor_music: computes a large set of spectral, time-domain, rhythm, tonal and high-level descriptors. The frame-wise descriptors are summarized by their statistical distribution and can be optionally output per frames. This extractor was designed for batch computations on large music collections and is the easiest way to get the most of music descriptors out of Essentia without programming. See detailed documentation.

  • essentia_streaming_extractor_freesound: similar extractor recommended for sound analysis. This extractor is used by Freesound in order to provide sound analysis API and search by similar sounds functionality. See detailed documentation.

  • essentia_standard_pitchdemo: extracts pitch for a monophonic/polyphonic signal using various pitch estimation algorithms.

  • essentia_streaming_predominantpitchmelodia: extracts pitch of a predominant melody using MELODIA algorithm.

  • essentia_streaming_beattracker_multifeature_mirex2013: extracts beat postions usign the multifeature beattracker algorithm.

  • essentia_streaming_mfcc: extracts MFCC frames and their statistical characterization.

  • essentia_standard_rhythmtransform: computes rhythm transform.

Given an audio file these extractors produce a yaml or json file with results.