warbleR (v1.1.24) now includes functions to detect signals using cross-correlation similar to those in the package monitoR. There is already a blog post on cross-correlation detection using monitoR. In this post I show how to do that with warbleR and compare its performance against that from monitoR.
First install the latest developmental version of warbleR and load other packages (the code will install the packages if missing):
devtools::install_github("maRce10/warbleR")
x <- c("warbleR", "monitoR", "microbenchmark", "ggplot2")
out <- lapply(x, function(y) {
if(!y %in% installed.packages()[,"Package"]) install.packages(y)
require(y, character.only = T)
})
and load the long-billed hermit sound that comes with warbleR:
#write files in temporal directory
setwd(tempdir()))
# load sound files and data
data(list = c("Phae.long1", "Phae.long2",
"Phae.long3", "Phae.long4", "selec.table"))
#write files to disk
writeWave(Phae.long1,"Phae.long1.wav")
Before getting into the functions let’s recall how detection works with monitoR:
- Create template
- Run cross-correlation
- Find peaks (detections above a trheshold)
# Create template
phae1T1 <- makeCorTemplate("Phae.long1.wav",
t.lim = c(lbh_selec_table$start[2], lbh_selec_table$end[2]),
wl = 300, ovlp = 90, frq.lim = c(1, 11), dens = 1,
name = "phae1T1", score.cutoff = 0.4, write.wav = FALSE)
# Run cross-correlation
cscoresPhae1 <- corMatch(survey = "Phae.long1.wav",
templates = phae1T1, parallel = TRUE,
show.prog = FALSE, time.source = "fileinfo")
# Find peaks
cdetectsPhae1 <- findPeaks(cscoresPhae1, parallel = TRUE)
This graph shows the detections:
# View results
plot(cdetectsPhae1, hit.marker="points")
Below is the exact same process using warbleR. Note that the template(s) and sound files to be used in the detections are indicated in the “compare.matrix” input. This is simply a matrix with the templates in the first column (“Phae.long1.wav-1” means selection “1” from sound file “Phae.long1.wav”) and the sound files in which to look for in the second column. The object ’lbh_selec_table’ is a selection table with all the annotation data needed to find templates:
# make matrix with templates and sound files
cm <- matrix(c("Phae.long1.wav-1", "Phae.long1.wav"), ncol = 2)
# Run cross correlation
xc.output <- xcorr(X = lbh_selec_table[1, ], output = "list",
compare.matrix = cm, path = tempdir())
# find peaks
pks <- find_peaks(xc.output = xc.output, path = tempdir())
# add frequency range (not required but used for plot)
pks$bottom.freq <- lbh_selec_table$bottom.freq[1]
pks$top.freq <- lbh_selec_table$top.freq[1]
And plot the detections:
spectrograms(X = pks, wl = 200, by.song = "sound.files",
collevels = seq(-80, 0, 5), xl = 3, pb = FALSE)
Compare warbleR and monitoR
We will use the function microbenchmark() from the package of the same name to test performance (i.e. how long it takes to run). To do this we first need to save both processes as individual functions:
# monitoR detection
detec_monitor <- function(x){
# Create template
phae1T1 <- makeCorTemplate("Phae.long1.wav",
t.lim = c(lbh_selec_table$start[2], lbh_selec_table$end[2]),
wl = 300, ovlp = 90, frq.lim = c(1, 11), dens = 1,
name = "phae1T1", score.cutoff = 0.4, write.wav = FALSE)
# Run cross-correlation
cscoresPhae1 <- corMatch(survey = "Phae.long1.wav",
templates = phae1T1, parallel = TRUE,
show.prog = FALSE, time.source = "fileinfo")
# find peaks
cdetectsPhae1 <- findPeaks(cscoresPhae1)
}
# warbleR detection
detec_warbleR <- function(x){
xc.output <- x_corr(X = lbh_selec_table[1, ], output = "list",
compare.matrix = matrix(c("Phae.long1.wav-1", "Phae.long1.wav"), ncol = 2),
path = tempdir(), pb = FALSE)
# find peaks
pks <- find_peaks(xc.output = xc.output, path = tempdir(),
pb = FALSE)
pks$bottom.freq <- lbh_selec_table$bottom.freq[1]
pks$top.freq <- lbh_selec_table$top.freq[1]
}
And now test performance (this can take a few minutes):
mb <- microbenchmark(detec_warbleR(), detec_monitor(), times = 10)
autoplot(mb)
# print results
mb
## Unit: milliseconds
## expr min lq mean
## detec_warbleR() 76.35316 100.0231 136.2479
## detec_monitor() 687.64700 879.5293 1009.3005
## median uq max neval cld
## 119.6610 140.7929 416.8854 100 a
## 927.2823 1031.3535 2224.5691 100 b
# mean time per function
mean_times <- tapply(mb$time, mb$expr, mean)
# how monitoR compares to warbleR
round(mean_times[2] / mean_times[1], 2)
## detec_monitor()
## 7.41
Detection with monitoR takes ~7 times longer. So a long process that may take a week in monitoR would take about a day using warbleR. Note that parallelization could further improve performance. This can be set with the argument ‘parallel’ in both x_corr() and find_peaks().
Session information
## R version 3.6.1 (2019-07-05)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 18.04.4 LTS
##
## Matrix products: default
## BLAS: /usr/lib/x86_64-linux-gnu/atlas/libblas.so.3.10.3
## LAPACK: /usr/lib/x86_64-linux-gnu/atlas/liblapack.so.3.10.3
##
## locale:
## [1] LC_CTYPE=es_ES.UTF-8
## [2] LC_NUMERIC=C
## [3] LC_TIME=es_CR.UTF-8
## [4] LC_COLLATE=es_ES.UTF-8
## [5] LC_MONETARY=es_CR.UTF-8
## [6] LC_MESSAGES=es_ES.UTF-8
## [7] LC_PAPER=es_CR.UTF-8
## [8] LC_NAME=C
## [9] LC_ADDRESS=C
## [10] LC_TELEPHONE=C
## [11] LC_MEASUREMENT=es_CR.UTF-8
## [12] LC_IDENTIFICATION=C
##
## attached base packages:
## [1] stats graphics grDevices utils
## [5] datasets methods base
##
## other attached packages:
## [1] ggplot2_3.3.0 microbenchmark_1.4-7
## [3] monitoR_1.0.7 warbleR_1.1.23
## [5] NatureSounds_1.0.2 knitr_1.28
## [7] seewave_2.1.5 tuneR_1.3.3
##
## loaded via a namespace (and not attached):
## [1] Rcpp_1.0.3 mvtnorm_1.0-11
## [3] lattice_0.20-40 prettyunits_1.0.2
## [5] ps_1.3.0 fftw_1.0-6
## [7] zoo_1.8-6 assertthat_0.2.1
## [9] rprojroot_1.3-2 digest_0.6.22
## [11] R6_2.4.1 backports_1.1.5
## [13] signal_0.7-6 evaluate_0.14
## [15] highr_0.8 pillar_1.4.2
## [17] rlang_0.4.1 curl_4.2
## [19] multcomp_1.4-12 rstudioapi_0.10
## [21] callr_3.3.2 Matrix_1.2-18
## [23] desc_1.2.0 devtools_2.2.1
## [25] splines_3.6.1 stringr_1.4.0
## [27] RCurl_1.98-1.1 munsell_0.5.0
## [29] proxy_0.4-23 compiler_3.6.1
## [31] xfun_0.12 pkgconfig_2.0.3
## [33] pkgbuild_1.0.6 tidyselect_0.2.5
## [35] tibble_2.1.3 dtw_1.21-3
## [37] codetools_0.2-16 crayon_1.3.4
## [39] dplyr_0.8.3 withr_2.1.2
## [41] MASS_7.3-51.4 bitops_1.0-6
## [43] grid_3.6.1 gtable_0.3.0
## [45] magrittr_1.5 scales_1.0.0
## [47] cli_1.1.0 stringi_1.4.6
## [49] pbapply_1.4-2 fs_1.3.1
## [51] remotes_2.1.0 testthat_2.3.0
## [53] ellipsis_0.3.0 sandwich_2.5-1
## [55] TH.data_1.0-10 rjson_0.2.20
## [57] tools_3.6.1 glue_1.3.1
## [59] purrr_0.3.3 processx_3.4.1
## [61] pkgload_1.0.2 parallel_3.6.1
## [63] survival_3.1-11 yaml_2.2.1
## [65] colorspace_1.4-1 sessioninfo_1.1.1
## [67] memoise_1.1.0 usethis_1.5.1