Sinusoidal and envelope modulation modeling of signals - a signal theoretic approach to acoustic events rendering

Abstract

This study investigates a signal theoretic approach to rendering acoustic events. Acoustic events modeling language (AEML) is an essential part of acoustic events rendering using an acoustic-data stream based on structured audio representation. This article mainly describes sinusoidal and envelope-modulation modeling for intelligible speech modification and reverberation signals rendering. The sinusoidal modeling is useful for constructing intelligible speech using only a few dominant components, and narrow-band envelopes such as 1/4- octave-band-speech envelopes are the key to representation of speech intelligibility. Envelope modulation modeling using the dominant sinusoidal carriers enables modification of the talker's pitch and speech-rate without sacrificing intelligibility. The narrowband envelope can be estimated by clustered linespectrum modeling (CLSM) based on the LSE-criterion in the frequency domain. Sinusoidal modeling with a decaying envelope is also a key technology for reverberation sound rendering based on the modal statistics of a reverberation field.