MarmosetCallClassification

Classification of marmoset vocalizations

The model gets 96% correct on the test set.

Performance

Performance of the TweetyNet model trained on marmoset vocalizations*:

Accuracy	Levenshtein	Loss	Segment Error Rate
0.9618	195.1	0.1217	0.3174

• Recorded and labeled by Daniel Y. Takahashi and colleauges. See Takahashi, D. Y., Fenley, A. R., Teramoto, Y., Narayanan, D. Z., Borjon, J. I., Holmes, P., & Ghazanfar, A. A. (2015). The developmental dynamics of marmoset monkey vocal production. Science, 349(6249), 734-738. for details.

Installation

1. Install vak version >= 0.8.
2. Download the marmoset call classfication model and extract the ZIP archive.

Use

1. Prepare directories and the configuration file. Either by creating the directories (relative to your working directory) in predict_example.toml (data/WAV, data/preprocessed and results/predictions), or by editing the configuration file predict.toml so that the paths point to the correct locations.

    [PREP]
    data_dir = "</path/to/the/WAV-files/to/run/the/model/on>"
    output_dir = "</path/to/where/the/organizing/CSV-file/should/be/stored>"
    spect_output_dir = "</path/to/where/the/spectrograms/should/be/stored>"
    [PREDICT]
    device = '<"cpu" or "cuda">'
    checkpoint_path = "</path/to/the/model/>TweetyNet/checkpoints/max-val-acc-checkpoint.pt"
    labelmap_path = "</path/to/the/model/>labelmap.json"
    spect_scaler_path = "</path/to/the/model/>StandardizeSpect"
    output_dir = "</dir/where/the/resulting/CSV-file/should/be/stored>"
    annot_csv_filename = "<name of the CSV-file with the predictions>.csv"

2. Pre-process the WAV-files: vak prep predict.toml
3. Predict: vak predict predict.toml. The resulting CSV file should be located and named according to annot_csv_filename above.
4. (Optional) Plot the predictions for manual inspection:

   python3 plot.py <annot_csv_filename> <spec_dir> <fig_dir> --plot_dur 10 --mark_onsets --mark_offsets