Abstract
Rodents’ ultrasonic vocalization (USV) provides useful information to assess their social behaviors. Despite of previous efforts for classifying subcategories of time-frequency patterns of USV syllables to associate with their functional relevances, detection of vocal elements from continuously recorded data have remained to be not well-optimized. We here propose a novel procedure for detecting USV segments in continuous sound data with background noises which were inevitably contaminated during observation of the social behavior. The proposed procedure utilizes a stable version of spectrogram and additional signal processing for better separation of vocal signals by reducing variation of the background noise. Our procedure also provides a precise time tracking of spectral peaks within each syllable. We showed that this procedure can be applied to a variety of USVs obtained from several rodent species. A performance test with an appropriate parameter set showed performance for detecting USV syllables than conventional methods.