SCMIR: A SuperCollider Music Information Retrieval Library
Nick Collins · University of Michigan Library Repository · 2011
The SuperCollider Music Information Retrieval (SCMIR) library is an extension set for the SuperCollider audio programming language that facilitates automatic analysis of audio files.The framework supports common music information retrieval technologies, including for batch processing across sound file collections.The library takes advantage of scsynth's Non-Real-Time mode and machine listening plug-ins for fast feature extraction, as well as SuperCollider language invocation of auxiliary native externals for some intensive calculations.Features can be normalized/standardized, grouped by detected beat and onset locations as well as arbitrary imposed segmentations.Similarity matrices support multiple distance metrics, novelty curve calculation through checkerboard kernel, and dynamic time warping best match path discovery.Applications include automatic structure analysis of pieces, inter-piece comparison, and as a front-end to machine learning operations via SC classes or Weka.1 There are also online web services offered by universities and companies (such as services through the EchoNest API), but these are limited in application for large audio databases, where work on a local machine is much more flexible and uninhibited by upload and download limits, and third party APIs not so open to researcher tweaking (it is impossible to modify the signal analysis routines provided by Tristan Jehan in EchoNest, for example).