Context
Audio analysis using machine learning methods for classification begins with an extraction of suitable features. Well-known methods for extraction of audio features are often re-implemented by many researchers working in the field of audio analysis. The low level feature extraction plugin API (FEAPI) provides a generic interface for audio feature extraction modules.

History
At the beginning of January 2005, Alexander Lerch posted a message on the music-ir mailing list with the idea (and a first draft) for a generic audio feature extraction plugin API, openly sollicitating feedback and collaboration from other R&D engineers working in this field. The idea was to make it an easy-to-use, open, platform-independent and very permissively licensed API that could be used as a common interface for all parties creating or using low-level audio feature extraction software modules.

Capabilities
The capabilities of FEAPI were defined as:

  • support for different and possibly varying sample rates of the extracted features
  • support for multiple independent instances of each plugin
  • support for multidimensional features
  • push-style processing of audio buffers (data source can be anything: files, live streams, ...)
  • support for sufficient timing information to allow synchronization of features with different sample rates
  • support for the calculation of multiple features in one plugin, if required by the developer
  • high probability of unique plugin identification by the host without a registration process

License
The FEAPI code is licensed under a BSD style license, which makes it usable in both open or closed source applications, commercial and non-commercial.

See also

  • Features (pattern recognition)
  • Feature extraction
  • Machine learning
  • Pattern recognition