An automatic detection algorithm for extracting the representative frequency of cetacean tonal sounds

Most studies on tonal sounds extract contour parameters from fundamental frequencies. The presence of harmonics and the frequency distribution of multiple tonal sounds have not been well researched. To investigate the occurrence and frequency modulation of cetacean tonal sounds, the procedure of detecting the instantaneous frequency bandwidth of tonal spectral peaks was integrated within the local-max detector to extract adopted frequencies. The adopted frequencies, considered the representative frequencies of tonal sounds, are used to find the presence of harmonics and overlapping tonal sounds. The utility and detection performance are demonstrated on acoustic recordings of five species from two databases. The recordings of humpback dolphins showed a 75% detection rate with a 5% false detection rate, and recordings from the MobySound archive showed an 85% detection rate with a 5% false detection rate. These detections were achieved in signal-to-noise ratios of -12 to 21 dB. The parameters that measured the distribution of adopted frequency, as well as the prominence of harmonics and overlaps, indicate that the modulation of tonal sounds varied among different species and behaviors. This algorithm can be applied to studies on cetacean communication signals and long-term passive acoustic monitoring. (C) 2013 Acoustical Society of America.