MIR How-Tos

Vienna University of Technology
Institute of Software Technology and Interactive Systems
Information & Software Engineering Group

Music Information Retrieval

[Topics] [Projects] [Downloads] [People] [Publications] [Press] [Events]

How-To Collection

Matlab Audio Feature Extraction

1. filelist

Create a filelist of the audio files you want. The files should be in subdirectories of one "base directory". To create a filelist you can do something similar like this:

Linux:
find . -name *.mp3 > filelist.txt

Windows:
dir /s /b *.mp3 > filelist.txt (?not approved yet!)

2. install helper applications

Matlab can read .wav and .au files by itself. For other file formats (like mp3 and ogg) you need external tools to decode the files.
You can find Windows versions of external tools here:

mp3: mpg123.exe or lame.exe (you need only one of these)
ogg: oggdec
Free Lossless Audio Codec (flac): flac installer
Monkey's Audio Codec (mac): mac installer

In Linux some libraries are probably already installed or can be installed by a search on the package name similar to the above names from a repository for your distribution.

3. set the feature extraction options

Copy rp_options.txt to a directory of your choice and adjust the parameters described in it.

4. start the feature extraction

Read the variants for the syntax for calling rp_extract from whithin MATLAB
Note: write rp_extract in lowercase letters.
Recommended syntax: Call

rp_extract('/path/to/rp_options.txt')
If you want to call the algorithm directly from command line through MATLAB use:
Linux:
matlab -r rp_extract\(\'/path/to/options-file\'\) -nosplash -nodesktop
For users logged in on the athena server, as a shorter variant of the direct command line call you can use:
rp_extract [path/to/]options_file

5. post processing

Finally the following output files are generated in the output directory you have set in rp_options.txt:

<projectname>.log ... the log file, which you can inspect, if there were errors
<projectname>.rp ... Rhythm Patterns feature vector file
<projectname>.rh ... Rhythm Histogram feature vector file
<projectname>.ssd ... Statistical Spectrum Descriptor feature vector file

The three feature vector files can be used as input vectors for the SOM-Toolbox or for classification.

top	last edited 20.07.2006 by Thomas Lidy