Figure 1.

Spectrograms of a 4 second segment of audio, containing a series of coughs, showing stages of algorithm; a) 512 sample Fast Fourier Transform of audio data is taken every 16 samples and positive half of spectrum is plotted vertically using intensity of colour as contribution of frequency to signal. This spectrogram is shown underneath the audio waveform. b) Median frequency is calculated for each point in spectrogram and shown here superimposed. c) Threshold value for median frequency used to generate a ‘mask’ indicating which audio is to be removed, indicated graphically here as the faded sections. d) Timings for cuts of audio data adjusted to ensure ‘attack’ of sounds is captured.