Fig. 3

TrackUSF accurately captures the majority of the manually detected USVs and enables their further characterization. A An example of a spectrogram showing a sequence of USVs, as defined manually (orange boxes) and by DeepSqueak (blue boxes). USFs detected by TrackUSF at two threshold levels are marked by the green (threshold = 2.7 a.u.) and pink (threshold = 1 a.u.) asterisks. B Mean percentage of manually defined USVs, overlapped by at least one USF, using five different thresholds (1, 1.5, 2.2, 2.7, and 3.5 a.u.), as well as by DeepSqueak-defined USVs (gray bar). C Mean percentage of coverage of the total duration of manually defined USVs by USFs for each of the various thresholds, as well as by DeepSqueak-defined USVs (gray bar). D Number of detected USFs for each audio clip plotted as a function of the number of the corresponding manually detected USVs, for each of the various thresholds. Each dot represents a distinct audio clip analyzed using a distinct threshold. Distinct colors represent distinct thresholds, as depicted in the legend. E Percentage of all detected USFs from all clusters, except for cluster 1 (noise), that were found to represent non-USV signals (false-positive detections) for each of the various thresholds, as well as for DeepSqueak. Each point represents a distinct audio clip. F Mean PSD profiles of calls emitted by C57BL/6J mice pairs detected either manually (orange), by DeepSqueak (blue), or by TrackUSF (green). The TrackUSF analysis used a threshold of 2.7 a.u., after scaling the PSD curve of each cluster to the number of USFs and summing the curves separately for C57BL/6J mice. G As is F, for the calls emitted by BalbC mice. H Normalized probability function of DeepSqueak-detected USVs according to their duration. I Normalized probability function of DeepSqueak-detected USVs according to their slope. J K-means clustering of DeepSqueak-detected USVs according to their length, slope, and principal frequency, separately for each mouse strain. All error bars represent SEM