Data-driven speech intelligibility prediction has been slow totake off. Datasets of measured speech intelligibility are scarce,and so current models are relatively small and rely on hand-picked features. Classical predictors based on psychoacousticmodels and heuristics are still the state-of-the-art. This workproposes a U-Net inspired fully convolutional neural networkarchitecture, NSIP, trained and tested on ten datasets to pre-dict intelligibility of time-domain speech. The architecture iscompared to a frequency domain data-driven predictor and tothe classical state-of-the-art predictors STOI, ESTOI, HASPIand SIIB. The performance of NSIP is found to be superior fordatasets seen in the training phase. On unseen datasets NSIPreaches performance comparable to classical predictors.

Many deep learning-based speech enhancement algorithms are designed to minimize the mean-square error (MSE) in some transform domain between a predicted and a target speech signal. However, optimizing for MSE does not necessarily guarantee high speech quality or intelligibility, which is the ultimate goal of many speech enhancement algorithms. Additionally, only little is known about the impact of the loss function on the emerging class of time-domain deep learning-based speech enhancement systems. We study how popular loss functions influence the performance of time-domain deep learning-based speech enhancement systems. First, we demonstrate that perceptually inspired loss functions might be advantageous over classical loss functions like MSE. Furthermore, we show that the learning rate is a crucial design parameter even for adaptive gradient-based optimizers, which has been generally overlooked in the literature. Also, we found that waveform matching performance metrics must be used with caution as they in certain situations can fail completely. Finally, we show that a loss function based on scale-invariant signal-to-distortion ratio (SI-SDR) achieves good general performance across a range of popular speech enhancement evaluation metrics, which suggests that SI-SDR is a good candidate as a general-purpose loss function for speech enhancement systems.

The majority of deep neural network (DNN) based speech enhancement algorithms rely on the mean-square error (MSE) criterion of short-time spectral amplitudes (STSA), which has no apparent link to human perception, e.g., speech intelligibility. Short-time objective intelligibility (STOI), a popular state-of-the-art speech intelligibility estimator, on the other hand, relies on linear correlation of speech temporal envelopes. This raises the question if a DNN training criterion based on envelope linear correlation (ELC) can lead to improved speech intelligibility performance of DNN-based speech enhancement algorithms compared to algorithms based on the STSA-MSE criterion. In this paper, we derive that, under certain general conditions, the STSA-MSE and ELC criteria are practically equivalent, and we provide empirical data to support our theoretical results. Furthermore, our experimental findings suggest that the standard STSA minimum-MSE estimator is near optimal, if the objective is to enhance noisy speech in a manner, which is optimal with respect to the STOI speech intelligibility estimator.

In this paper we propose a Deep Neural Network (D NN) based Speech Enhancement (SE) system that is designed to maximize an approximation of the Short-Time Objective Intelligibility (STOI) measure. We formalize an approximate-STOI cost function and derive analytical expressions for the gradients required for DNN training and show that these gradients have desirable properties when used together with gradient based optimization techniques. We show through simulation experiments that the proposed SE system achieves large improvements in estimated speech intelligibility, when tested on matched and unmatched natural noise types, at multiple signal-to-noise ratios. Furthermore, we show that the SE system, when trained using an approximate-STOI cost function performs on par with a system trained with a mean square error cost applied to short-time temporal envelopes. Finally, we show that the proposed SE system performs on par with a traditional DNN based Short- Time Spectral Amplitude (STSA) SE system in terms of estimated speech intelligibility. These results are important because they suggest that traditional DNN based STSA SE systems might be optimal in terms of estimated speech intelligibility.

We propose a novel deep learning training criterion, named permutation invariant training (PIT), for speaker independent multi-talker speech separation, commonly known as the cocktail-party problem. Different from the multi-class regression technique and the deep clustering (DPCL) technique, our novel approach minimizes the separation error directly. This strategy effectively solves the long-lasting label permutation problem, that has prevented progress on deep learning based techniques for speech separation. We evaluated PIT on the WSJ0 and Danish mixed-speech separation tasks and found that it compares favorably to non-negative matrix factorization (NMF), computational auditory scene analysis (CASA), and DPCL and generalizes well over unseen speakers and languages. Since PIT is simple to implement and can be easily integrated and combined with other advanced techniques, we believe improvements built upon PIT can eventually solve the cocktail-party problem.

In this paper, we study aspects of single microphone speech enhancement (SE) based on deep neural networks (DNNs). Specifically, we explore the generalizability capabilities of state-of-the-art DNN-based SE systems with respect to the background noise type, the gender of the target speaker, and the signal-to-noise ratio (SNR). Furthermore, we investigate how specialized DNN-based SE systems, which have been trained to be either noise type specific, speaker specific or SNR specific, perform relative to DNN based SE systems that have been trained to be noise type general, speaker general, and SNR general. Finally, we compare how a DNN-based SE system trained to be noise type general, speaker general, and SNR general performs relative to a state-of-the-art short-time spectral amplitude minimum mean square error (STSA-MMSE) based SE algorithm. We show that DNN-based SE systems, when trained specifically to handle certain speakers, noise types and SNRs, are capable of achieving large improvements in estimated speech quality (SQ) and speech intelligibility (SI), when tested in matched conditions. Furthermore, we show that improvements in estimated SQ and SI can be achieved by a DNN-based SE system when exposed to unseen speakers, genders and noise types, given a large number of speakers and noise types have been used in the training of the system. In addition, we show that a DNN-based SE system that has been trained using a large number of speakers and a wide range of noise types outperforms a state-of-the-art STSA-MMSE based SE method, when tested using a range of unseen speakers and noise types. Finally, a listening test using several DNN-based SE systems tested in unseen speaker conditions show that these systems can improve SI for some SNR and noise type configurations but degrade SI for others.

In this paper, we propose the utterance-level permutation invariant training (uPIT) technique. uPIT is a practically applicable, end-to-end, deep-learning-based solution for speaker independent multitalker speech separation. Specifically, uPIT extends the recently proposed permutation invariant training (PIT) technique with an utterance-level cost function, hence eliminating the need for solving an additional permutation problem during inference, which is otherwise required by frame-level PIT. We achieve this using recurrent neural networks (RNNs) that, during training, minimize the utterance-level separation error, hence forcing separated frames belonging to the same speaker to be aligned to the same output stream. In practice, this allows RNNs, trained with uPIT, to separate multitalker mixed speech without any prior knowledge of signal duration, number of speakers, speaker identity, or gender. We evaluated uPIT on the WSJ0 and Danish two- and three-talker mixed-speech separation tasks and found that uPIT outperforms techniques based on nonnegative matrix factorization and computational auditory scene analysis, and compares favorably with deep clustering, and the deep attractor network. Furthermore, we found that models trained with uPIT generalize well to unseen speakers and languages. Finally, we found that a single model, trained with uPIT, can handle both two-speaker, and three-speaker speech mixtures.

In this paper we propose to use a state-of-the-art Deep Recurrent Neural Network (DRNN) based Speech Enhancement (SE) algorithm for noise robust Speaker Verification (SV). Specifically, we study the performance of an i-vector based SV system, when tested in noisy conditions using a DRNN based SE front-end utilizing a Long Short-Term Memory (LSTM) architecture. We make comparisons to systems using a Non-negative Matrix Factorization (NMF) based front-end, and a Short-Time Spectral Amplitude Minimum Mean Square Error (STSA-MMSE) based front-end, respectively. We show in simulation experiments that a male-speaker and text-independent DRNN based SE front-end, without specific a priori knowledge about the noise type outperforms a text, noise type and speaker dependent NMF based front-end as well as a STSA-MMSE based front-end in terms of Equal Error Rates for a large range of noise types and signal to noise ratios on the RSR2015 speech corpus.