Reverberation Time Estimation for Speech Processing Applications

The reverberation time (RT) is an important measure for the characterization of reverberant environments, which can be determined by different acoustical measurement methods. However, they use mostly a special measurement setup and dedicated excitation signals, which is inapplicable for most speech processing algorithms requiring knowledge about the RT. In this contribution, a new method is devised which allows to estimate the RT from noisy observations. It is based on a maximum likelihood (ML) estimation which is derived from a statistical model of the sound decay in reverberant and noisy enclosures. This allows to determine the RT from a measured sound decay (or room impulse response) despite the presence of noise. It is also shown, how the ML estimator can be used to determine the RT blindly from a noisy and reverberant speech signal. This blind RT estimation can be employed for the enhancement of noisy and reverberant speech signals.