This is a collection of R scripts to analyze the robustness of audio descriptors to different encodings and analysis parameters, as performed for instance in this paper:
- J. Urbano, D. Bogdanov, P. Herrera, E. Gómez and X. Serra, "What is the Effect of Audio Quality on the Robustness of MFCCs and Chroma Features?", in 15th ISMIR Conference, 2014.
How to run
robustness.R as the main entry point from
> cd R > Rscript robustness.R usage: Rscript robustness.R <command> <tool> <track-length> <srate> <descriptor> [<path>] command command to run (see below). tool name of the tool used to compute descriptors, eg. 'essentia'. track-length length of the tracks used to compute descriptors, eg '30-60'. srate sampling rate of the tracks used to compute descriptors, eg. '44100'. descriptor name of the descriptor to analyze, eg. 'lowlevel.mfcc.mean'. path optional path to all data (see below), defaults to '../data'. Available commands are (in expected order of execution): extract extract data specific of the descriptor. indicators compute robustness indicators. distributions describe the distributions of the indicators. boxplots boxplot the distributions of the indicators. variance decompose in variance components. -- all run all commands (extract, indicators, distributions, variance). help show this message. Expected file structure of descriptors: path/descriptors/<tool>/<track-length>/<srate>/<codec>/<brate>/<param1>-...-<paramN>/<genre>/file Output file structure: path/results/<tool>/<track-length>/<srate>/<descriptor>/<command>.txt
The following R packages are required: