In real-world environments, speech signals are often corrupted by ambient noises during their acquisition, leading to degradation of quality and intelligibility of the speech for a listener. As one of the central topics in the speech processing area, speech enhancement aims to recover clean speech from such a noisy mixture. Many traditional speech enhancement methods designed based on statistical signal processing have been proposed and widely used in the past. However, the performance of these methods was limited and thus failed in sophisticated acoustic scenarios. Over the last decade, deep learning as a primary tool to develop data-driven information systems has led to revolutionary advances in speech enhancement. In this context, speech enhancement is treated as a supervised learning problem, which does not suffer from issues faced by traditional methods. This supervised learning problem has three main components: input features, learning machine, and training target. In this thesis, various deep learning architectures and methods are developed to deal with the current limitations of these three components. First, we propose a serial hybrid neural network model integrating a new low-complexity fully-convolutional convolutional neural network (CNN) and a long short-term memory (LSTM) network to estimate a phase-sensitive mask for speech enhancement. Instead of using traditional acoustic features as the input of the model, a CNN is employed to automatically extract sophisticated speech features that can maximize the performance of a model. Then, an LSTM network is chosen as the learning machine to model strong temporal dynamics of speech. The model is designed to take full advantage of the temporal dependencies and spectral correlations present in the input speech signal while keeping the model complexity low. Also, an attention technique is embedded to recalibrate the useful CNN-extracted features adaptively. Through extensive comparative experiments, we show that the proposed model significantly outperforms some known neural network-based speech enhancement methods in the presence of highly non-stationary noises, while it exhibits a relatively small number of model parameters compared to some commonly employed DNN-based methods. Most of the available approaches for speech enhancement using deep neural networks face a number of limitations: they do not exploit the information contained in the phase spectrum, while their high computational complexity and memory requirements make them unsuited for real-time applications. Hence, a new phase-aware composite deep neural network is proposed to address these challenges. Specifically, magnitude processing with spectral mask and phase reconstruction using phase derivative are proposed as key subtasks of the new network to simultaneously enhance the magnitude and phase spectra. Besides, the neural network is meticulously designed to take advantage of strong temporal and spectral dependencies of speech, while its components perform independently and in parallel to speed up the computation. The advantages of the proposed PACDNN model over some well-known DNN-based SE methods are demonstrated through extensive comparative experiments. Considering that some acoustic scenarios could be better handled using a number of low-complexity sub-DNNs, each specifically designed to perform a particular task, we propose another very low complexity and fully convolutional framework, performing speech enhancement in short-time modified discrete cosine transform (STMDCT) domain. This framework is made up of two main stages: classification and mapping. In the former stage, a CNN-based network is proposed to classify the input speech based on its utterance-level attributes, i.e., signal-to-noise ratio and gender. In the latter stage, four well-trained CNNs specialized for different specific and simple tasks transform the STMDCT of noisy input speech to the clean one. Since this framework is designed to perform in the STMDCT domain, there is no need to deal with the phase information, i.e., no phase-related computation is required. Moreover, the training target length is only one-half of those in the previous chapters, leading to lower computational complexity and less demand for the mapping CNNs. Although there are multiple branches in the model, only one of the expert CNNs is active for each time, i.e., the computational burden is related only to a single branch at anytime. Also, the mapping CNNs are fully convolutional, and their computations are performed in parallel, thus reducing the computational time. Moreover, this proposed framework reduces the latency by %55 compared to the models in the previous chapters. Through extensive experimental studies, it is shown that the MBSE framework not only gives a superior speech enhancement performance but also has a lower complexity compared to some existing deep learning-based methods.

This book covers the state-of-the-art in deep neural-network-based methods for noise robustness in distant speech recognition applications. It provides insights and detailed descriptions of some of the new concepts and key technologies in the field, including novel architectures for speech enhancement, microphone arrays, robust features, acoustic model adaptation, training data augmentation, and training criteria. The contributed chapters also include descriptions of real-world applications, benchmark tools and datasets widely used in the field. This book is intended for researchers and practitioners working in the field of speech processing and recognition who are interested in the latest deep learning techniques for noise robustness. It will also be of interest to graduate students in electrical engineering or computer science, who will find it a useful guide to this field of research.

With the proliferation of mobile devices and hearing devices, including hearing aids and cochlear implants, there is a growing and pressing need to design algorithms that can improve speech intelligibility without sacrificing quality. Responding to this need, Speech Enhancement: Theory and Practice, Second Edition introduces readers to the basic pr

Speech enhancement (SE) aims to improve the speech quality of the degraded speech. Recently, researchers have resorted to deep-learning as a primary tool for speech enhancement, which often features deterministic models adopting supervised training. Typically, a neural network is trained as a mapping function to convert some features of noisy speech to certain targets that can be used to reconstruct clean speech. These methods of speech enhancement using neural networks have been focused on the estimation of spectral magnitude of clean speech considering that estimating spectral phase with neural networks is difficult due to the wrapping effect. As an alternative, complex spectrum estimation implicitly resolves the phase estimation problem and has been proven to outperform spectral magnitude estimation. In the first contribution of this thesis, a fully convolutional neural network (FCN) is proposed for complex spectrogram estimation. Stacked frequency-dilated convolution is employed to obtain an exponential growth of the receptive field in frequency domain. The proposed network also features an efficient implementation that requires much fewer parameters as compared with conventional deep neural network (DNN) and convolutional neural network (CNN) while still yielding a comparable performance. Consider that speech enhancement is only useful in noisy conditions, yet conventional SE methods often do not adapt to different noisy conditions. In the second contribution, we proposed a model that provides an automatic "on/off" switch for speech enhancement. It is capable of scaling its computational complexity under different signal-to-noise ratio (SNR) levels by detecting clean or near-clean speech which requires no processing. By adopting information maximizing generative adversarial network (InfoGAN) in a deterministic, supervised manner, we incorporate the functionality of SNR-indicator into the model that adds little additional cost to the system. We evaluate the proposed SE methods with two objectives: speech intelligibility and application to automatic speech recognition (ASR). Experimental results have shown that the CNN-based model is applicable for both objectives while the InfoGAN-based model is more useful in terms of speech intelligibility. The experiments also show that SE for ASR may be more challenging than improving the speech intelligibility, where a series of factors, including training dataset and neural network models, would impact the ASR performance.

This work addresses this problem in the short-time Fourier transform (STFT) domain. We divide the general problem into five basic categories depending on the number of microphones being used and whether the interframe or interband correlation is considered. The first category deals with the single-channel problem where STFT coefficients at different frames and frequency bands are assumed to be independent. In this case, the noise reduction filter in each frequency band is basically a real gain. Since a gain does not improve the signal-to-noise ratio (SNR) for any given subband and frame, the noise reduction is basically achieved by liftering the subbands and frames that are less noisy while weighing down on those that are more noisy. The second category also concerns the single-channel problem. The difference is that now the interframe correlation is taken into account and a filter is applied in each subband instead of just a gain. The advantage of using the interframe correlation is that we can improve not only the long-time fullband SNR, but the frame-wise subband SNR as well. The third and fourth classes discuss the problem of multichannel noise reduction in the STFT domain with and without interframe correlation, respectively. In the last category, we consider the interband correlation in the design of the noise reduction filters. We illustrate the basic principle for the single-channel case as an example, while this concept can be generalized to other scenarios. In all categories, we propose different optimization cost functions from which we derive the optimal filters and we also define the performance measures that help analyzing them.

We live in a noisy world! In all applications (telecommunications, hands-free communications, recording, human-machine interfaces, etc) that require at least one microphone, the signal of interest is usually contaminated by noise and reverberation. As a result, the microphone signal has to be "cleaned" with digital signal processing tools before it is played out, transmitted, or stored. This book is about speech enhancement. Different well-known and state-of-the-art methods for noise reduction, with one or multiple microphones, are discussed. By speech enhancement, we mean not only noise reduction but also dereverberation and separation of independent signals. These topics are also covered in this book. However, the general emphasis is on noise reduction because of the large number of applications that can benefit from this technology. The goal of this book is to provide a strong reference for researchers, engineers, and graduate students who are interested in the problem of signal and speech enhancement. To do so, we invited well-known experts to contribute chapters covering the state of the art in this focused field.

The above and most other speech separation systems operate on the magnitude response of noisy speech and use the noisy phase during signal reconstruction. This occurs because it is believed that the phase spectrum is unimportant for speech enhancement. More recent studies, however, reveal that phase is important for perceptual quality. We present an approach that concurrently enhances the magnitude and phase spectra by operating in the complex domain. We start by introducing the complex ideal ratio mask (cIRM), which has real and imaginary components. A DNN is used to jointly estimate these components of the cIRM. Evaluation results demonstrate that the proposed system substantially improves perceptual quality over recent approaches in noisy environments. Along with background noise, room reverberation is commonly encountered in real environments. The performance of many speech processing applications is severely degraded when both noise and reverberation are present. We propose to simultaneously perform dereverberation and denoising with the cIRM. First, we redefine the cIRM for reverberant and noisy environments. A DNN is then trained to estimate it. The complex mask removes the interference caused by noise and reverberation, and results in better predicted speech quality and intelligibility.