usv

A Python library and tools for classifying animal ultrasonic vocalizations using supervised machine learning. Developed for use with recordings from Avisoft-Recorder but is compatible with recordings from any software that records to WAV.

Integrates a number of 3rd-party libraries, including Orange for machine learning, Yaafe for feature extraction, and SoX for segmentation.

Requirements

• Python >= 2.7
• Orange
• Yaafe
• SoX (included in bin/)

Installation

After installing the dependencies, grab and install package:

git clone https://github.com/sloria/usv.git
cd usv/
python setup.py install

Add usv/scripts to your path:

export PATH=$PATH:$PWD/usv/scripts

OR add the following to ~/.bash_profile:

export PATH=$PATH:~/path/to/usv/scripts

Configure settings (IMPORTANT)

Copy the settings-dist.txt file to settings.txt (which won't be added to version control).

cp settings-dist.txt settings.txt

Edit the following settings in settings.txt:

• train_src: The absolute path includes the training data files
  • Example: /Users/johndoe/mouse_calls
• train_dest: Directory where training data sets will be saved to.
  • Example: /Users/johndoe/usv/trainsets

Full documentation to come...

Classification

First, cd to the directory containing the .wav files. Then, resample and segment all the audio files in the directory using

sqk.py -S

You can also isolate a specified audio channel. For example,

sqk.py -Sm 2

isolates channel 2 before reampling and segmenting.

Then, while still in the same directory, classify all the files in the folder using

sqk.py -C

Again, you can specify an audio channel, e.g.,

sqk.py -Cm 2

for channel 2.

The output is a .tab file with the classification counts for each file as well as the total classfication counts and proportions. The .tab file can be opened with a plain text editor or a spreadsheet application like Microsoft Excel.

View more options using sqk --help.

Full documentation to come...