Deep Neural Network Based Speech Separation Optimizing an Objective Estimator of Intelligibility for Low Latency Applications

Abstract

Mean square error (MSE) has been the preferred choice as loss function in the current deep neural network (DNN) based speech separation techniques. In this paper, we propose a new cost function with the aim of optimizing the extended short time objective intelligibility (ESTOI) measure. We focus on applications where low algorithmic latency (≤ 10 ms) is important. We use long short-term memory networks (LSTM) and evaluate our proposed approach on four sets of two-speaker mixtures from extended Danish hearing in noise (HINT) dataset. We show that the proposed loss function can offer improved or at par objective intelligibility (in terms of ESTOI) compared to an MSE optimized baseline while resulting in lower objective separation performance (in terms of the source to distortion ratio (SDR)). We then proceed to propose an approach where the network is first initialized with weights optimized for MSE criterion and then trained with the proposed ESTOI loss criterion. This approach mitigates some of the losses in objective separation performance while preserving the gains in objective intelligibility.

Components loss for neural networks in mask-based speech enhancement

“…The maximization of STOI [40] during training is also the target in several publications [31][32][33][34][35][36]. In [33], Kolbcek et al derive a differentiable approximation of STOI, which considers the frequency selectivity of the human ear, for the training of a mask-based speech enhancement DNN.…”

“…The parameters of the deep learning architectures are then optimized by minimizing the MSE between the inferred results and their corresponding targets. In reality, optimization of the MSE loss in training does not guarantee any perceptual quality of the speech component and of the residual noise component, respectively, which leads to limited performance [27][28][29][30][31][32][33][34][35][36]. This effect is even more evident when the level of the noise component is significantly higher than that of the speech component in some regions of the noisy speech spectrum, which explains the bad performance at lower SNR conditions when training with MSE.…”

“…A more straightforward direction is to utilize the shorttime objective intelligibility (STOI) [40] and the perceptual evaluation of speech quality (PESQ) [41] metrics as a loss function, which could be used to optimize for speech intelligibility and speech quality, respectively, during training [31][32][33][34][35][36]. Using STOI as an optimization criterion has been studied in [33,35,36].…”

Improving speech enhancement by focusing on smaller values using relative loss

“…Venkataramani and Smaragdis [31] use an objective function optimisation model that is equivalent to the signal-to-distortion ratio (SDR) measure, which can significantly improve the SDR scores of SE. Later, Kolbaek et al [32] and Naithani et al [33] propose loss functions for optimising the shorttime objective intelligibility (STOI) and the extended STOI, respectively, and Kim et al [34] present a denoising framework with the goal of joint SDR and perceptual evaluation of speech quality (PESQ) optimisation.…”

“…[32] and Naithani et al . [33] propose loss functions for optimising the short‐time objective intelligibility (STOI) and the extended STOI, respectively, and Kim et al . [34] present a denoising framework with the goal of joint SDR and perceptual evaluation of speech quality (PESQ) optimisation.…”