Deep Audio Segmenter (DAS)

DAS automatically annotates animal vocalizations in raw audio recordings using a deep neural network. It can be used through a graphical user interface, from the terminal, or from Python scripts.

Installation

On Windows and Linux:

conda create -n das -c conda-forge --strict-channel-priority python=3.14 pyside6 ffmpeg "numcodecs<0.16" uv -y

On macOS, select the OpenBLAS build that avoids an OpenMP conflict with PyTorch:

conda create -n das -c conda-forge --strict-channel-priority python=3.14 pyside6 ffmpeg "numcodecs<0.16" "libopenblas=*=*pthreads*" uv -y

Then, on all platforms:

conda activate das
uv pip install das --torch-backend=auto
das version

See the installation guide for CPU- and GPU-specific options.

Users who need the TensorFlow backend should follow the TensorFlow installation instructions for the final TensorFlow-backed release (0.32.13).

Documentation

See the DAS documentation for the complete user guide:

• The quick-start tutorials for flies and birds cover manual annotation, network training, and generating new annotations.
• Use the graphical user interface.
• Use DAS from the terminal or Python scripts.

If you have questions, feedback, or find a bug, please open an issue.

Citation

Please cite DAS as:

Elsa Steinfath, Adrian Palacios, Julian Rottschäfer, Deniz Yuezak, Jan Clemens (2021). Fast and accurate annotation of acoustic signals with deep neural networks. eLife

Acknowledgements

The following packages were modified and integrated into DAS:

• Keras implementation of TCN models modified from keras-tcn (in das.models.tcn)
• Trainable STFT layer implementation modified from kapre (in das.models.kapre)

See the corresponding source directories for the original READMEs.