In these exercises the main goal was to test and tune the sofware I wrote many years ago, https://github.com/ecoz2/ecoz2.
The following files constitute the initial set of 10 labelled files used as a basis for the exercises described here:
Wave file selection file (TSV)
======================== ===================
HBSe_20151023T122324.wav 20151023T122324.txt
HBSe_20151121T040102.wav 20151121T040102.txt
HBSe_20151207T070326.wav 20151207T070326.txt
HBSe_20151228T103639.wav 20151228T103639.txt
HBSe_20161101T153358.wav 20161101T153358.txt
HBSe_20161207T115528.wav 20161207T115528.txt
HBSe_20170107T085150.wav 20170107T085150.txt
HBSe_20170116T054541.wav 20170116T054541.txt
HBSe_20170424T102157.wav 20170424T102157.txt
These original files are not captured in this repository.
However, using a separate tool, all unit selections
were extracted as individual WAV files and committed into this repository.
They are located under corresponding class name subdirectories at data/signals/*:
$ ls data/signals
ascending_grunt groan modulated_cry_(D)
ascending_moan groan_+_purr modulated_moan
ascending_moan? groan_+purr modulated_moan_+_upsweep
ascending_moan_(D) growl modulated_moans_(F)
ascending_moan_+_purr growl+gurgle pulse
ascending_shriek grumble pulses
ascending_shriek,_higher grunt purr
ascending_shriek_(D) grunt_(long) purr_(D)
bark grunt_(s) purr_+_modulated_moan
bark? grunt_(short) scream
bark_(D) grunt_+_modulated_moan screech_?
bellow grunt_echo short_moan
broadband_burst grunts sigh
creek gurgle siren
croak gurgle? snort
cry gurgle?_ascending__ thwop
descending_grunt gurgle_+_descending_shriek trill
descending_moan gurgle_+_trill trill?
descending_moan_(F) gurgle_+modulated_moan trill_(D)
descending_shriek long_grunt trumpet?
descending_shriek? low_yaps violin?
descending_shriek_(D) modulated_cry wops
Each subdirectory contains all WAV files corresponding to the associated
class name, which is the value of the Description column in all
selection files, except in the case of 20151207T070326.txt, where the
selected column was Description (Dunlop & Fournet).
(Spaces have been converted to underscores in all class names.)
As an example, these are the selection WAV files for the croak class:
$ ls data/signals/croak
from_HBSe_20151228T103639__333.30273_334.11978.wav
from_HBSe_20151228T103639__345.19217_346.69003.wav
from_HBSe_20151228T103639__356.40396_359.39975.wav
from_HBSe_20151228T103639__367.97855_371.31476.wav
from_HBSe_20151228T103639__380.1414_383.13718.wav
from_HBSe_20151228T103639__391.79767_395.0658.wav
from_HBSe_20151228T103639__405.1632_408.0228.wav
from_HBSe_20151228T103639__417.58368_419.21774.wav
from_HBSe_20151228T103639__430.50308_431.8648.wav
from_HBSe_20151228T103639__441.06293_443.17358.wav
from_HBSe_20151228T103639__453.8636_455.7019.wav
from_HBSe_20170116T054541__387.63336_389.26135.wav
The name of each extracted file indicates the original WAV file name and the corresponding begin and end times.
The exercises described here are:
whale1: Only uses selections from one of the song files.whale10: Uses selections from the set of 10 song files.
To reproduce these exercises you will need to have the ECOZ2 executables
(lpc, vq.learn, hmm.classify, etc.) on your machine.
See https://github.com/ecoz2/ecoz2.
Note that only the various "report" files (.rpt) generated during
codebook and HMM model training are committed in this repo
(but no other generated files).