Figure 7.

Simulation of the breathing-sound recognition with dynamically changing analysis parameters. Predicted breathing-sound classification when the model was free to choose the breathing-sound parameter (sound level, power spectrum or roughness) that provides the strongest predictions (strongest deviations from chance performance) for a given comparison of a test sound with each of the three training sounds. All sounds were filtered to match either the human audiogram (a) or the vampire-bat audiogram (b). While the simulation based on the human audiogram does not yield improved predictions of the classification of the test sounds recorded under physical strain, this simulation approach when combined with the sounds as they are weighted by the vampire-bat audiogram results in qualitatively correct classifications even of the test sounds recorded under physical strain.

Gröger and Wiegrebe BMC Biology 2006 4:18   doi:10.1186/1741-7007-4-18
Download authors' original image