BASS-dB: the Blind Audio Source Separation evaluation database
Evaluation toolbox
Performance measures
The performance of a source separation algorithm on a given mixture is evaluated using the Source-to-Distortion Ratio (SDR), the Source-to-Interferences Ratio (SIR) and the Source-to-Artifacts Ratio (SAR) in deciBels for each estimated source. When a single performance measure is needed, these quantities are averaged between sources to obtain the mean values SDRm, SIRm and SARm. These performance measures work in an evaluation framework where the original sources are available for comparison. They take into account the gain/filtering indeterminacies of Blind Source Separation algorithms and account separately for the amount of residual crosstalk (from non-target sources) and artifacts (such as water pipe noise) in each estimated source. More details are provided in
BSS_EVAL Toolbox
These criteria are implemented in a MATLAB toolbox called BSS_EVAL.
Download the most recent version of the toolbox (including the user guide).
Read more about the toolbox.