Software used in study of noise from ship traffic.
The raw data from the hydrophone on the Tiburon research pier, stored as 15-minute WAVE fles, is automatically archived on Ingrid.sfsu.edu, at /usr/local/data1/rtc/wav/ , typical file name rtc120703192504.wav . JPEG's and MP3's are calculated and posted on the web site, bayacoustics.sfsu.edu.
A series of special files is prepared for computer recognition of ship sounds. Each raw-data file is passed through successive "cleaning" operations:
- First dehumming: a frequency line in a band [80-120 Hz] and its harmonics are replaced by adjacent "noise."
- Deglitching: Short, narrow peaks (spikes, or glitches) in the intensity are replaced by adjacent "noise." The results are stored as files rdg120703192504.wav .
- Second and third dehumming: The dehumming procedure is repeated twice after deglitching. Only "signifcant" frequency lines are removed. The resulting WAVE files are stored at ingrid%/usr/local/data1/rtc/rdh as rdh120703192504.wav files.
- The program which carries out this cleaning is mkdgdh.m, using dehumming program dhm15h.m, and deglitching program dgl15.m.
The identification of ship sounds is done by searching the .rdh files for peaks in the intensity. The WAVE time series is squared, decimated by x10, and smoothed by x5. The resulting time series is searched for peaks. The algorithm consists (roughly) of finding the highest point and its half-power points; fitting with a Gaussian plus constant background; replacing the peak by adjacent background; and iterating until the waveform is flat. Program sman45.m calls getshp45.m and gauspeak.m . The programs with "45" in the name work with three joined 15-minute time series, only keeping signals peaking in the central 15 minutes. This permits detecting sounds which overlap 15-minute boundaries.