Human hearing more capable than models predicted

Ars Technica : Humans hear by using their ears to detect vibrations and their brain to translate those vibrations into nerve inpulses. Most models of human hearing assume that the sound measurements made by the ear and brain are linear. The result is a direct correlation between the abilities to accurately determine the frequency and the duration of a sound. Such linear models are used in modern audio compression algorithms to save space by removing presumably inaudible tones. To fine-tune the model, Jacob Oppenheim and Marcelo Magnasco of Rockefeller University in New York tested the ability of subjects to determine both the frequency and duration of so-called Fourier-limited sound pulses. Such pulses are pairings of specific frequencies and shortest possible durations that together result in the minimum amount of uncertainty possible according to linear models. The researchers found that across the board the test subjects outperformed by a factor of 10 the linear model’s predictions for ability to determine the sound pulses’ frequency, duration, or both. Better understanding of the way that humans detect sound may produce audio compression algorithms that improve sound quality and don’t cut out parts of the sound that people can actually hear.