plot.cond: Plot several DET plots from a single trial list, analyzed per...

In davidavdav/ROC: Compute structures to compute ROC and DET plots and metrics for 2-class classifiers.

Plot several DET plots from a single trial list, analyzed per condition

Description

This function plots DET curves for all conditions in a multi-condition trial set, such as NIST SRE-2008. An overall DET curve, equalizing trial counts, is also plotted.

Usage

plot.cond(x, cond, nr = 1, equalize = T, ...)

Arguments

x	A data.frame of class sre
cond	A list of factors specifying the conditions
nr	The number (color) of the first line.
equalize	Do we equalize trial counts for the overall DET curve?
...	further parameters for det.plot()

Details

For each condition determined by the list of factors, a separate DET will be plotted in a new color. The overall condition is plotted in black, as a last curve so that it will be “on top of” the other curves. You may find yourself soon to be running out of sensible colors.

It is possible to specify equalize=FALSE purely for demonstration purposes, we don't believe this makes very much sense.

Value

A table with a summary of the main detection performance metrics for each condition, and averaged over all conditions (but equalized for trial counts).

Author(s)

David van Leeuwen

References

1. David A. van Leeuwen, “Overal performance metrics for multi-condition Speaker Recognition Evaluations,” Proc. Interspeech, Brighton, September 2009, pp 908–911.

2. Milou van Dijk, Rosemary Orr, David van der Vloed and David van Leeuwen, “A human benchmark for automatic speaker recognition”, Proc.\ of the 1st International Conference Biometric Technologies in Forensic Science, Nijmegen 2013.

See Also

cond.table, plot.det

Examples

## load data from the TNO submission to the NIST Speaker Recognition Evaluation 2008, see [1]
data(tno.2008)
## plot for all sensible conditions of NIST SRE 2008
plot.cond(tno.2008, list(mtype, ttype, tmic))
## another eval, EVALITA, the Radboud University Nijmegen submission
data(ru.2009)
setDCF("evalita")
## we can plot fol all train/test conditions in one go...
plot.cond(ru.2009, list(mcond, tcond))
## or look at gender effect, or channel
x <- subset(ru.2009, mcond=="TC6" & tcond=="TS2")
plot.cond(x, gender)
plot.cond(x, channel)
## Finally, some human speaker recognition data, with ordered factor levels, collected by Milou [2]
data(milou)
head(milou)
plot.cond(milou, cl, equalize=F, where="ll", main="Naive human speaker recognition by difficulty class")

davidavdav/ROC documentation built on Sept. 8, 2023, 2:39 p.m.