das_unsupervised 0.6.0

Tools for unsupervised classification of acoustic signals.

Tools for unsupervised classification of acoustic signals

DAS-unsupervised provides tools for pre-processing acoustic signals for unsupervised classification:

• extract waveforms or spectrograms of acoustic events from a recording
• normalize the duration, center frequency, amplitude, or sign of waveform/spectrograms

Unsupervised classification itself is performed using existing libraries:

• dimensionality reduction: umap
• clustering: hdbscan or scikit-learn
• NEW: interactive embedding and visualization with marimo.

Can be used in combination with DAS, a deep learning based method for the supervised annotation of acoustic signals.

Install via conda and uv

conda create -y -n das_unsupervised -c conda-forge python=3.13 uv
conda activate das_unsupervised
uv pip install das_unsupervised

Demos

Illustration of the workflow and the method using vocalizations from:

• flies
• mice
• birds
• NEW: interactive zebra finch

Acknowledgements

Code from the following open source packages was modified and integrated into das-unsupervised:

• avgn (Sainburg et al. 2020)
• noisereduce
• fly pulse classifier (Clemens et al. 2018)

Data sources:

• flies: David Stern (Stern, 2014)
• mice: data provided by Kurt Hammerschmidt (Ivanenko et al. 2020)
• birds: Bengalese finch song repository (Nicholson et al. 2017), Zebra finch song repository (Goffinet et al. 2021)

References

1. T Sainburg, M Thielk, TQ Gentner (2020) Latent space visualization, characterization, and generation of diverse vocal communication signals. Biorxiv . https://doi.org/10.1101/870311

2. J Clemens, P Coen, F Roemschied, T Perreira, D Mazumder, D Aldorando, D Pacheco, M Murthy (2018) Discovery of a New Song Mode in Drosophila Reveals Hidden Structure in the Sensory and Neural Drivers of Behavior. Current Biology 28, 2400–2412.e6 (2018). https://doi.org/10.1016/j.cub.2018.06.011

3. D Stern (2014). Reported Drosophila courtship song rhythms are artifacts of data analysis. BMC Biology

4. A Ivanenko, P Watkins, MAJ van Gerven, K Hammerschmidt, B Englitz (2020) Classifying sex and strain from mouse ultrasonic vocalizations using deep learning. PLoS Comput Biol 16(6): e1007918. https://doi.org/10.1371/journal.pcbi.1007918

5. D Nicholson, JE Queen, S Sober (2017). Bengalese finch song repository. https://doi.org/10.6084/m9.figshare.4805749.v5