Speech Dereverberation Using Non-Negative Convolutive Transfer Function and Spectro-Temporal Modeling

Abstract

This paper presents two single-channel speech dereverberation methods to enhance the quality of speech signals that have been recorded in an enclosed space. For both methods, the room acoustics are modeled using a non-negative approximation of the convolutive transfer function (N-CTF), and to additionally exploit the spectral properties of the speech signal, such as the low-rank nature of the speech spectrogram, the speech spectrogram is modeled using non-negative matrix factorization (NMF). Two methods are described to combine the N-CTF and NMF models. In the first method, referred to as the integrated method, a cost function is constructed by directly integrating the speech NMF model into the N-CTF model, while in the second method, referred to as the weighted method, the N-CTF and NMF based cost functions are weighted and summed. Efficient update rules are derived to solve both optimization problems. In addition, an extension of the integrated method is presented, which exploits the temporal dependencies of the speech signal. Several experiments are performed on reverberant speech signals with and without background noise, where the integrated method yields a considerably higher speech quality than the baseline N-CTF method and a state-of-the-art spectral enhancement method. Moreover, the experimental results indicate that the weighted method can even lead to a better performance in terms of instrumental quality measures, but that the optimal weighting parameter depends on the room acoustics and the utilized NMF model. Modeling the temporal dependencies in the integrated method was found to be useful only for highly reverberant conditions.

Multichannel Online Dereverberation Based on Spectral Magnitude Inverse Filtering

“…(Q+Õ−1) ≤ IÕ and thusÕ ≥Q −1 I−1 , (19) has an exact solution and (20) can reach zero. Otherwise, (19) is a least square problem, and only an approximate solution can be achieved.…”

“…Note that with the superscript (p) removed, h and J denote the stationary filter and the (stationary) random variable for the squared error, respectively. At frame p, the instantaneous filtering process in (19) and the squared error (20) are a random instance of the stationary system. At frame p, the adaptive update uses the gradient of the instantaneous error J (p) at the previous estimation point h (p−1) , i.e.…”

Time-Frequency Masking in the Complex Domain for Speech Dereverberation and Denoising

“…In [21] and [35], the T-F magnitude response of the RIR is estimated. Another approach uses the RIR magnitude response and nonnegative matrix factorization (NMF) to remove reverberation [27]. A two-stage algorithm for enhancing reverberant speech is described by Wu and Wang [43], where the first stage estimates an inverse filter and the second stage uses spectral subtraction to minimize long-term reverberation.…”

Single-Channel Online Enhancement of Speech Corrupted by Reverberation and Noise

“…Most single-channel speech dereverberation techniques can be classified into inverse filtering [10], [11], nonlinear mapping [12], spectral enhancement [6], [13], [14] and probabilistic model-based methods [15]- [17]. Inverse filtering methods typically try to reconstruct the original signal by designing an inverse filter for the Room Impulse Response (RIR).…”