Room impulse response inference from the coherence properties of speech and music signals

Abstract

Analysis of the coherence properties of the harmonic partials of speech and music signals recorded in an acoustic space provides information about the impulse response of the space. When an acoustic signal is filtered by a space, the autocorrelation of individual partials and the cross-correlation between pairs of partials provides information about the frequency-dependent reverberation time of the space. The relationship between the coherence time of the original, dry, signal and the reverberation time of the acoustic space determines the information that is retrievable in this “blind” approach. Recovering information about the acoustic space can be done within bounds, given only minimal assumptions or knowledge of the signal. Specifically, the original signal is assumed to contain a set of harmonically related partials, with mutually correlated amplitude and phase modulations and that the coherence time of the signal is known. These modulations are modelled as a narrow band pseudo random process. This is a reasonable assumption based on the physics of sound generation in classes of realistic sources, such as wind musical instruments or the human vocal tract.