The BOSARIS Toolkit provides MATLAB code for calibrating, fusing and evaluating scores from (automatic) binary classifiers. It was developed to provide solutions for automatic speaker recognition, but we envision that much of the code will have wider applicability for other biometric and/or forensics problems, where the calibration of likelihood-ratios is of interest. The main features of the toolkit are:
- The Normalized Bayes Error-Rate Plot, which analyses likelihood ratio calibration over a wide range of operating points. These plots also help in judging the adequacy of the sizes of calibration and evaluation databases.
- ROC, ROCCH and DET plots.
- Efficient algorithms to compute these plots for very large score files.
- Logistic regression solution for fusing multiple sub-systems.
- Logistic regression, or PAV/ROCCH calibration solutions for mapping scores to likelihood-ratios.
- A fast logistic regression optimizer.
- An efficient, binary, platform-independent (HDF5) score file format, which facilitates interoperability with other tools.
More details are available in:
New for SRE'12: Utility for simple to compound LLR transformation
This may be helpful to NIST SRE'12 participants. BOSARIS fusion and calibration outputs simple LLRs. For some SRE'12 test conditions you will need compound LLRs. The MATLAB function llrTrans_simple2compound.m may be useful for this purpose. Also consult the SRE'12 page for more information and resources.
The toolkit has an object-oriented API, for which it needs a recent verson of MATLAB
. We have verified that it runs on MATLAB version R2008a and later versions.
- if !agree with License agreement return end
- Read the user guide, to see what the toolkit does and why.
- Download: The toolkit is available in both zipped (.zip) and tar gzipped (.tgz) versions. The contents are identical. See attachments at the bottom of this page.
- Go to: getting started, which will tell you again to read the user guide, to add the code to the MATLAB path and to run the demo.