Dynamic Precedence Effect Modeling for Source Separation in Reverberant Environments

Abstract

Abstract-Reverberation continues to present a major problem for sound source separation algorithms. However, humans demonstrate a remarkable robustness to reverberation and many psychophysical and perceptual mechanisms are well documented. The precedence effect is one of these mechanisms; it aids our ability to localise sounds in reverberation. Despite this, relatively little work has been done on incorporating the precedence effect into automated source separation. Furthermore, no work has been carried out on adapting a precedence model to the acoustic conditions under test and it is unclear whether such adaptation, analogous to the perceptual Clifton effect, is even necessary. Hence, this study tests a previously proposed binaural separation/precedence model in real rooms with a range of reverberant conditions. The precedence model inhibitory time constant and inhibitory gain are varied in each room in order to establish the necessity for adaptation to the acoustic conditions. The study concludes that adaptation is necessary and can yield significant gains in separation performance. Furthermore, it is shown that the Initial Time Delay Gap and the Direct-to-Reverberant Ratio are important factors when considering this adaptation.

Two-Microphone Dereverberation for Automatic Speech Recognition of Polish

“…We modified parameters g and α P of the precedence model to achieve an optimal performance on the target system (Polish ASR) during the preliminary tests. It was found that fixed values other than those suggested by Hummersone (Hummersone et al, 2010) can provide better results. In our optimisation both parameters were increased for the 16 kHz signals, gain factor by 6% to g = 0.825, and the time constant by 100% to α P = 12.5.…”

“…It is based on the perceptual emphasis of the first wave front and has been implemented to both dereverberate and separate speech signals (Palomaki et al, 2004). We tested and implemented a precedence model developed by (Hummersone et al, 2010).…”

“…In an auditory implementation (Hummersone et al, 2010), the magnitude output is as in Eq. (4) with p = 1, time constant α P , h(t) = At exp(−t/α P ) with normalisation factor A, and X(t, ω) being the Hilbert envelope of the Gammatone filterbank signals.…”

An EM algorithm for joint source separation and diarisation of multichannel convolutive speech mixtures

“…Each source signal was a 27s speech signal made by concatenating utterances chosen from the TIMIT database [21] (one different speaker for each source). As mixing filters, we used binaural room impulse responses from [22] having RT 60 ≈ 0.68s. We generated two types of mixtures: Mix-DC where all sources are emitting continuously.…”

Spatial and coherence cues based time-frequency masking for binaural reverberant speech separation

“…The precedence effect is also applied and modeled as a low pass filter with an inhibitory gain and time constant to mitigate the reverberant energy from the signal and to improve the separation performance in [17]. However, the model parameters are required to be estimated based on the room properties which are not always known as a priori.…”