Efficient calculation of a physiologically-motivated representation for sound

Abstract

An algorithm is proposed which calculates a computationally efficient approximation of a certain physiologically-motivated representation for sound, called the summary autocorrelation function. This representation has been found very useful in several tasks, such as sound separation, multiple period estimation, and computational auditory scene analysis. However, it has been computationally too complex for most practical applications. The relatively fast algorithm described here proposes only an approximation of the summary autocorrelation function, but the achieved precision is likely to be good enough for most applications.