Fig. 1

The TrackUSF pipeline. A All audio clips of a given experiment are pooled and analyzed together. B Each audio clip is divided into 6-ms fragments. C All fragments then pass a 15-kHz high-pass filter, and those that contain signals exceeding a predetermined power threshold (USFs) are collected. D The power spectrum between 15 and 100 kHz of each USFs is converted to 16 Mel-frequency cepstral coefficients (MFCCs). E MFCCs of all USFs are then analyzed together using a 3-dimensional (3D) T-distributed Stochastic Neighbor Embedding (t-SNE) for dimensionality reduction and visualization. F Clusters are defined, either manually or using DBSCAN automatic clustering algorithm. G Example of further analyses of the results enabled by TrackUSF. H Also enabled is the examination of the USFs on the spectrogram of their corresponding audio clip