Intelligent Sound Processing

Home of the music information retrieval toolbox

Features

The Intelligent Sound Processing Toolbox contains a great range of functions related to music information retrieval. Its major features are:

Functions for feature extraction, such as mel-frequency cepstral coefficients, chromagram etc.
A complete test framework which is also prepared for most standard audio collections such as the ISMIR 2004 genre classification training set and ballroom data, the covers80 and artist20 sets, and a number of custom tests based on MIDI synthesis.
Functions that makes distributing computations among multiple machines easy, even without the Distributed Computing Toolbox. Of course, the test framework makes full use of these.
Tools for manipulating and synthesizing MIDI files, thus making it easy to make synthetic tests that show specific properties of algorithms.
A fast implementation of Dan Ellis' instantaneous frequency-based chromagram, and reasonably fast Gaussian mixture model tools.
Distributed under the GNU General Public License version 2. If you need a less restrictive license, contact us.
Does not use any of MATLAB's object oriented features.

Documentation

Documentation of the included functions can be found here. Be sure to have a look at the demonstrations.

Download

The ISP toolbox is still in beta, so features might change without notice. Download the full version here (152 MB), or the reduced version without sound fonts and source code here (14 MB). Older versions are available here.

About

The ISP Toolbox is developed as part of the Intelligent Sound Project and is maintained by Jesper Højvang Jensen. Other contributors from the Intelligent Sound Project are:

Anders Meng
Elvis Ramonovic
Kaare Brandt Petersen
Mads Græsbøll Christensen
Sigurdur Sigurdsson
Tue Lehn-Schiøler

Acknowledgements

The original versions of isp_mp3read, isp_ifchromagram, isp_ifptrack and isp_ifgram are all written by Dan Ellis.
The files isp_midiread and isp_midiwrite are extended versions of the corresponding files from the Midi Toolbox.
The file isp_mfccvb uses functions from Mike Brookes' VOICEBOX toolbox.
This toolbox uses code from LAME, MAD, MF2T/T2MF, MPlayer, and TiMidity++.
The included Fluid Release 3 sound font was made by Frank Wen, and the test mp3 is from ccMixter.

Changelog

2010-02-09
: Thanks to Dan Ellis and Adrian Weller, the chromagram functions now also compile under Windows.
2009-07-28
: Added the isp_rhythdemo.m script and reorganized some of isp_tirhythm.m.
2008-08-14
: Bug in the distribution functions caused by the change of random number generator in Matlab 7.0sp3 fixed. Memory leak in chromagram function fixed thanks to Dan Ellis.
2008-06-29
: Minor bugfixes.
2008-06-02
: The beta of the ISP Toolbox v. 2 was released.
2006-04-03
: The Intelligent Sound Processing toolbox R1 is now available for download. Features: MFCC short-term features, multi-variate Auto-Regressive long-term feature modeling, maximum likelihood and correlation based fundamental frequency estimators, and Gaussian mixture model-based classification.
2006-03-22: On April 1 2006, the first version of the Intelligent Sound Processing toolbox for MATLAB will be released and made available for download on this homepage.
2006-02-02: This Intelligent Sound project server (isound.kom.auc.dk) is now up and running. It will host the Intelligent Sound Processing (ISP) toolbox for MATLAB.