A Priori SNR Estimation Based on a Recurrent Neural Network for Robust Speech Enhancement

Abstract

Speech enhancement under highly non-stationary noise conditions remains a challenging problem. Classical methods typically attempt to identify a frequency-domain optimal gain function that suppresses noise in noisy speech. These algorithms typically produce artifacts such as "musical noise" that are detrimental to machine and human understanding, largely due to inaccurate estimation of noise power spectra. The optimal gain function is commonly referred to as the ideal ratio mask (IRM) in neural-network-based systems, and the goal becomes estimation of the IRM from the short-time Fourier transform amplitude of degraded speech. While these data-driven techniques are able to enhance speech quality with reduced artifacts, they are frequently not robust to types of noise that they had not been exposed to in the training process. In this paper, we propose a novel recurrent neural network (RNN) that bridges the gap between classical and neural-network-based methods. By reformulating the classical decision-directed approach, the a priori and a posteriori SNRs become latent variables in the RNN, from which the frequency-dependent estimated likelihood of speech presence is used to update recursively the latent variables. The proposed method provides substantial enhancement of speech quality and objective accuracy in machine interpretation of speech.

Weighted Speech Distortion Losses for Neural-Network-Based Real-Time Speech Enhancement

“…During experiments, we notice that even though our systems trained on MSE (e.g. row 4 in Table 1) could achieve similar objective measures compared to those trained on the proposed weighted losses (12), the corresponding subjective quality of systems trained on the weighted loss is a lot better. The most noticeable improvement of systems trained on our loss functions, especially with small α, is that the estimated gain function is much more frequency-selective than systems trained on regular MSE, resulting in higher noise suppression, especially at high SNRs.…”

“…In this work, we study real-time speech enhancement with recurrent neural network (RNN). Recent works involving RNNs demonstrated promising results [10], even at very low signal-to-noise ratio (SNR) scenarios [11,12].…”

“…Methods along this line of thought include learning multiple objectives from heterogeneous features [16,17,18], jointly optimizing the final goal and its sub-targets (e.g. speech-presence probability) [10,12], and directly optimizing towards an objective measure of speech quality or intelligibility [19,20]. The latter seems a promising way to improve objective quality, although both models have to incorporate the standard MSE due to the band limitation of each objective measure.…”

Deep learning for minimum mean-square error approaches to speech enhancement

“…Unlike previous a priori SNR estimators, the proposed estimators do not require a noise estimator. Recently, a recurrent neural network (RNN) was used to aid the DD approach in a priori SNR estimation [15]. The proposed estimators differ by directly estimating the a priori SNR.…”

“…The proposed a priori SNR estimators significantly outperform the previous a priori SNR estimation methods. Evaluating the results in [15], the RNN-assisted DD approach (a deep learing-based a priori SNR estimator) could only outperform the DD approach at higher SNR levels (5 dB and greater for signal-to-distortion ratio (SDR)). Here, the ResLSTM and ResBLSTM a priori SNR estimators significantly outperform the DD approach for all conditions.…”