AMVOC

AMVOC analyzes ultrasonic vocalizations (USVs) in mice to detect, classify, and characterize vocal repertoires for studies of neural mechanisms and human speech-related disorders.

Key Features:

• Deep Unsupervised Learning: Employs a deep, unsupervised learning approach to discover vocalization types without manual labels.
• High Accuracy in Detection: Demonstrates high accuracy in detecting USVs against hand-annotated ground truth and retains performance in noisy recordings.
• Online and Offline Modes: Supports online (real-time) and offline analysis, enabling application to experiments including free behavior monitoring.

Scientific Applications:

• Neurobiology of communication: Provides quantitative data on mouse USVs to investigate neural mechanisms underlying vocal behavior.
• Modeling speech-related disorders: Facilitates studies relating mouse vocalization patterns to human speech-related disorders.
• Behavioral experiments: Enables real-time monitoring and offline analysis of vocalizations during free behavior experiments.

Methodology:

AMVOC uses an unsupervised deep learning framework based on convolutional autoencoders to extract latent features from USVs, which are then clustered and analyzed without predefined labels to explore vocal repertoire space.