Noise Induced Hearing Enhancement: Clinical and Machine Learning Studies

The addition of a certain amount of background noise has been shown to improve the hearing of pure-tone sounds. The underlying mechanism of this counter-intuitive phenomenon is believed to be stochastic resonance, where the threshold level of a pure-tone signal is lowered by adding an appropriate amount of background noise. Here, to investigate whether background noise can aid the hearing of more complex sound such as the human voice, we perform hearing tests with Korean syllables, and find that syllable recognition is enhanced with noise for rarely used syllables. Similar tests with a machine-learning speech recognition shows that the same enhancement arises only when the system is insufficiently trained, corresponding to hearing a rare syllable. The overall phenomenon looks similar to the stochastic resonance but any successful model should explain that the enhancement arises in perceptual processes as it depends on the level of training and the syllables.