evaluate_sad: evaluation of speech detection tools
In gobbios/avutils: various utilities for DiViMe tools and audio processing

Description Usage Arguments Value Examples

evaluation of speech detection tools

1
2
3

evaluate_sad(test, reference, nsegments = 1, duration = NULL,
  from = NULL, to = NULL, overlap_res = NULL, doplot = TRUE,
  raw_results = FALSE, ...)

`test`	character, path to file with annotations in either ELAN (.eaf) or DiViMe (.rttm) format
`reference`	character, path to reference file with annotations in either ELAN (.eaf) or DiViMe (.rttm) format
`nsegments`	numeric, the number of segments in which the file is to be split (represented as different lines in output graphic). Default is `1`, i.e. the entire file is considered as one segment.
`duration`	numeric, optional info about the duration of the audio. At its default `NULL`, the end of the last annotation is taken as the duration
`from`	numeric, starting point for selecting a subset of time
`to`	numeric, starting point for selecting a subset of time
`overlap_res`	numeric, the time resolution for calculating overlap, by default `NULL`, i.e. no overlap is calculated, see `annotation_overlap`
`doplot`	logical, should the plot actually be produced (default is `TRUE`)
`raw_results`	logical, should the raw results be returnd, i.e. the values per 'sample' (by default `FALSE`)
`...`	additional parameters for elan files (which tiers/annotations to ignore), see `collapse_tiers`

a data.frame with proportions of agreement between the two SADs

test <- system.file("noisemesSad_synthetic_speech.rttm", package = "avutils")
reference <- system.file("synthetic_speech.eaf", package = "avutils")
# no overlap calculation
evaluate_sad(test = test, reference = reference)
# fairly crude time resolution for overlap calculation
evaluate_sad(test = test, reference = reference, overlap_res = 10)
# slightly higher time resolution
evaluate_sad(test = test, reference = reference, overlap_res = 100)
# even more fine-grained
evaluate_sad(test = test, reference = reference, overlap_res = 1000)
# even more fine-grained and file split into multiple segments
evaluate_sad(test = test, reference = reference, nsegments = 10, overlap_res = 100)