davidavdav/sretools.R: Compute error metrics and make DET plots for NIST-like Speaker Recognition Evaluations

This package contains a set of routines that can compute various performance measures for NIST style speaker recognition evaluation system output. The routines in this package allow - trial key info sore SRE 2006, 2008 and EVALITA 2009 - flexible plotting of DET curves - calculation of EER, minCdet, Cdet, minCllr and Cllr - adding legends to the plot in the order of the EER - plotting Applied Probability of Error (APE)-plots, with proper legend - plotting of Normalized Bayes Error rate (NBE) plots - merging of meta-information to the scores, and manipulation of these tables - conditioning trials on meta-information - weighting trials based on a chosen conditioning, before calculation of DET parameters and performance metrics. This is pretty new, although George Doddington's DET_tools.pm has similar capability.

install.packages("remotes")
remotes::install_github("davidavdav/sretools.R")