David Nicholson- vak: neural network models for acoustic behavior | SciPy 2023

Neural network models for analyzing acoustic behavior in animals, using the VOC framework and exploring techniques for dimensionality reduction, model selection, and performance evaluation.

Key takeaways

Neural networks can be used to analyze acoustic behavior in animals, such as bird song or bat calls.
The speaker uses the VOC framework, a neural network framework, to analyze acoustic behavior.
VOC provides a single framework with multiple models and makes it easy to experiment with different models.
The speaker has developed a product for quality assurance of machine learning models.
The VOC framework has been used in other papers and has been shared with the community.
The speaker is working on a core library that will make it easier to share and use the models.
The framework is open source and can be used by anyone.
The speaker is using NumPy, pandas, and scikit-learn in their work.
The speaker is using a parametric UMAP model to reduce the dimensionality of the audio data.
The speaker is using a neural network to automatically segment the audio data.
The speaker is using a syllable error rate to evaluate the model’s performance.
The speaker is using a learning curve to understand how much training data is needed.
The speaker is working on an ablation experiment to understand the importance of different parts of the model.
The speaker is using a probabilistic suffix tree model to model canary song.
The speaker is using a neural network to embed the audio data into a lower-dimensional space.
The speaker is using a neural network to classify the audio data.
The speaker is working on a project to analyze the neural activity of canaries while they are singing.

David Nicholson- vak: neural network models for acoustic behavior | SciPy 2023

More talks