vak

automated annotation of animal vocalizations

vak is a library for researchers studying animal vocalizations. It automates annotation of vocalizations, using artificial neural networks. By annotation, we mean something like the example of annotated birdsong shown below:

You give vak training data in the form of audio or spectrogram files with annotations, and then vak helps you train neural network models and use the trained models to predict annotations for new files.

Installation

Short version:

$ pip install vak

For the long version detail, please see: https://vak.readthedocs.io/en/latest/get_started/installation.html

Usage

Training models to segment and label vocalizations

Currently the easiest way to work with vak is through the command line.

You run it with config.toml files, using one of a handful of commands.

For more details, please see the "autoannotate" tutorial here:
https://vak.readthedocs.io/en/latest/tutorial/autoannotate.html

Data and folder structures

To train models, you provide training data in the form of audio or spectrograms files, and annotations for those files.

Spectrograms and labels

The package can generate spectrograms from .wav files or .cbin audio files. It can also accept spectrograms in the form of Matlab .mat or Numpy .npz files. The locations of these files are specified in the config.toml file.

The annotations are parsed by a separate library, crowsetta, that aims to handle common formats like Praat textgrid files, and enable researchers to easily work with formats they may have developed in their own labs. For more information please see:
https://crowsetta.readthedocs.io/en/latest/
https://github.com/NickleDave/crowsetta

Preparing training files

It is possible to train on any manually annotated data but there are some useful guidelines:

• Use as many examples as possible - The results will just be better. Specifically, this code will not label correctly syllables it did not encounter while training and will most probably generalize to the nearest sample or ignore the syllable.
• Use noise examples - This will make the code very good in ignoring noise.
• Examples of syllables on noise are important - It is a good practice to start with clean recordings. The code will not perform miracles and is most likely to fail if the audio is too corrupt or masked by noise. Still, training with examples of syllables on the background of cage noises will be beneficial.

Predicting annotations for audio

You can predict annotations for audio files by creating a config.toml file with a [PREDICT] section.
For more details, please see the "autoannotate" tutorial here: https://vak.readthedocs.io/en/latest/tutorial/autoannotate.html

Support / Contributing

Currently we are handling support through the issue tracker on GitHub:
https://github.com/NickleDave/vak/issues
Please raise an issue there if you run into trouble.
That would be a great place to start if you are interested in contributing, as well.

Citation

If you use vak for a publication, please cite its DOI:

License

is here.

Misc

"Why this name, vak?"

It has only three letters, so it is quick to type, and it wasn't taken on pypi yet. Also I guess it has something to do with speech. "vak" rhymes with "squawk" and "talk".

Does your library have any poems?

Yes.