Given the recent surge in developments of deep learning, this paper provides a review of the state-of-the-art deep learning techniques for audio signal processing. Speech, music, and environmental sound processing are considered side-by-side, in order to point out similarities and differences between the domains, highlighting general methods, problems, key references, and potential for cross fertilization between areas. The dominant feature representations (in particular, log-mel spectra and raw waveform) and deep learning models are reviewed, including convolutional neural networks, variants of the long short-term memory architecture, as well as more audio-specific neural network models. Subsequently, prominent deep learning application areas are covered, i.e., audio recognition (automatic speech recognition, music information retrieval, environmental sound detection, localization and tracking) and synthesis and transformation (source separation, audio enhancement, generative models for speech, sound, and music synthesis). Finally, key issues and future questions regarding deep learning applied to audio signal processing are identified.

Deep Learning for Audio Signal Processing

In this paper, we propose a convolutional recurrent neural network for joint sound event localization and detection (SELD) of multiple overlapping sound events in three-dimensional (3-D) space. The proposed network takes a sequence of consecutive spectrogram time frames as input and maps it to two outputs in parallel. As the first output, the sound event detection (SED) is performed as a multi-label classification task on each time frame producing temporal activity for all the sound event classes. As the second output, localization is performed by estimating the 3-D Cartesian coordinates of the direction-of-arrival (DOA) for each sound event class using multi-output regression. The proposed method is able to associate multiple DOAs with respective sound event labels and further track this association with respect to time. The proposed method uses separately the phase and magnitude component of the spectrogram calculated on each audio channel as the feature, thereby avoiding any method- and array-specific feature extraction. The method is evaluated on five Ambisonic and two circular array format datasets with different overlapping sound events in anechoic, reverberant, and real-life scenarios. The proposed method is compared with two SED, three DOA estimation, and one SELD baselines. The results show that the proposed method is generic and applicable to any array structures, robust to unseen DOA values, reverberation, and low SNR scenarios. The proposed method achieved a consistently higher recall of the estimated number of DOAs across datasets in comparison to the best baseline. Additionally, this recall was observed to be significantly better than the best baseline method for a higher number of overlapping sound events.

Sound Event Localization and Detection of Overlapping Sources Using Convolutional Recurrent Neural Networks

Lacking of adaptation to various array imperfections is an open problem for most high-precision direction-of-arrival (DOA) estimation methods. Machine learning-based methods are data-driven, they do not rely on prior assumptions about array geometries, and are expected to adapt better to array imperfections when compared with model-based counterparts. This paper introduces a framework of the deep neural network to address the DOA estimation problem, so as to obtain good adaptation to array imperfections and enhanced generalization to unseen scenarios. The framework consists of a multitask autoencoder and a series of parallel multilayer classifiers. The autoencoder acts like a group of spatial filters, it decomposes the input into multiple components in different spatial subregions. These components thus have more concentrated distributions than the original input, which helps to reduce the burden of generalization for subsequent DOA estimation classifiers. The classifiers follow a one-versus-all classification guideline to determine if there are signal components near preseted directional grids, and the classification results are concatenated to reconstruct a spatial spectrum and estimate signal directions. Simulations are carried out to show that the proposed method performs satisfyingly in both generalization and imperfection adaptation.

/pdf/direction-of-arrival-estimation-based-on-deep-neural-5d6ze1o10h.pdf

Direction-of-Arrival Estimation Based on Deep Neural Networks With Robustness to Array Imperfections

Supervised learning-based methods for source localization, being data driven, can be adapted to different acoustic conditions via training and have been shown to be robust to adverse acoustic environments. In this paper, a convolutional neural network (CNN) based supervised learning method for estimating the direction of arrival (DOA) of multiple speakers is proposed. Multi-speaker DOA estimation is formulated as a multi-class multi-label classification problem, where the assignment of each DOA label to the input feature is treated as a separate binary classification problem. The phase component of the short-time Fourier transform (STFT) coefficients of the received microphone signals are directly fed into the CNN, and the features for DOA estimation are learnt during training. Utilizing the assumption of disjoint speaker activity in the STFT domain, a novel method is proposed to train the CNN with synthesized noise signals. Through experimental evaluation with both simulated and measured acoustic impulse responses, the ability of the proposed DOA estimation approach to adapt to unseen acoustic conditions and its robustness to unseen noise type is demonstrated. Through additional empirical investigation, it is also shown that with an array of M microphone our proposed framework yields the best localization performance with M -1 convolution layers. The ability of the proposed method to accurately localize speakers in a dynamic acoustic scenario with varying number of sources is also shown.

Multi-Speaker DOA Estimation Using Deep Convolutional Networks Trained With Noise Signals

This paper proposes a deep neural network for estimating the directions of arrival (DOA) of multiple sound sources. The proposed stacked convolutional and recurrent neural network (DOAnet) generates a spatial pseudo-spectrum (SPS) along with the DOA estimates in both azimuth and elevation. We avoid any explicit feature extraction step by using the magnitudes and phases of the spectrograms of all the channels as input to the network. The proposed DOAnet is evaluated by estimating the DOAs of multiple concurrently present sources in anechoic, matched and unmatched reverberant conditions. The results show that the proposed DOAnet is capable of estimating the number of sources and their respective DOAs with good precision and generate SPS with high signal-to-noise ratio.

Direction of Arrival Estimation for Multiple Sound Sources Using Convolutional Recurrent Neural Network

A Speaker Localization algorithm based on Neural Networks for multi-room domestic scenarios is proposed in this paper. The approach is fully data-driven and employs a Neural Network fed by GCC-PHAT (Generalized Cross Correlation Phase Transform) Patterns, calculated by means of the microphone signals, to determine the speaker position in the room under analysis. In particular, we deal with a multi-room case study, in which the acoustic scene of each room is influenced by sounds emitted in the other rooms. The algorithm is tested against the home recorded DIRHA dataset, characterized by multiple wall and ceiling microphone signals for each room. In particular, we focused on the speaker localization problem in two distinct neighbouring rooms. We assumed the presence of an Oracle multi-room Voice Activity Detector (VAD) in our experiments. A three-stage optimization procedure has been adopted to find the best network configuration and GCC-PHAT Patterns combination. Moreover, an algorithm based on Time Difference of Arrival (TDOA), recently proposed in literature for the addressed applicative context, has been considered as term of comparison. As result, the proposed algorithm outperforms the reference one, providing an average localization error, expressed in terms of RMSE, equal to 525 mm against 1465 mm. Concluding, we also assessed the algorithm performance when a real VAD, recently proposed by some of the authors, is used. Even though a degradation of localization capability is registered (an average RMSE equal to 770 mm), still a remarkable improvement with respect to the state of the art performance is obtained.

A neural network based algorithm for speaker localization in a multi-room environment

A novel end-to-end binaural sound localisation approach is proposed which estimates the azimuth of a sound source directly from the waveform. Instead of employing hand-crafted features commonly employed for binaural sound localisation, such as the interaural time and level difference, our end-to-end system approach uses a convolutional neural network (CNN) to extract specific features from the waveform that are suitable for localisation. Two systems are proposed which differ in the initial frequency analysis stage. The first system is auditory-inspired and makes use of a gammatone filtering layer, while the second system is fully data-driven and exploits a trainable convolutional layer to perform frequency analysis. In both systems, a set of dedicated convolutional kernels are then employed to search for specific localisation cues, which are coupled with a localisation stage using fully connected layers. Localisation experiments using binaural simulation in both anechoic and reverberant environments show that the proposed systems outperform a state-of-the-art deep neural network system. Furthermore, our investigation of the frequency analysis stage in the second system suggests that the CNN is able to exploit different frequency bands for localisation according to the characteristics of the reverberant environment.

/pdf/end-to-end-binaural-sound-localisation-from-the-raw-waveform-4m57xaor57.pdf

End-to-end Binaural Sound Localisation from the Raw Waveform

Localizing speakers in multiple rooms by using Deep Neural Networks

This paper focuses on Voice Activity Detectors (VAD) for multi-room domestic scenarios based on deep neural network architectures. Interesting advancements are observed with respect to a previous work. A comparative and extensive analysis is lead among four different neural networks (NN). In particular, we exploit Deep Belief Network (DBN), Multi-Layer Perceptron (MLP), Bidirectional Long Short-Term Memory recurrent neural network (BLSTM) and Convolutional Neural Network (CNN). The latter has recently encountered a large success in the computational audio processing field and it has been successfully employed in our task. Two home recorded datasets are used in order to approximate real-life scenarios. They contain audio files from several microphones arranged in various rooms, from whom six features are extracted and used as input for the deep neural classifiers. The output stage has been redesigned compared to the previous author's contribution, in order to take advantage of the networks discriminative ability. Our study is composed by a multi-stage analysis focusing on the selection of the features, the network size and the input microphones. Results are evaluated in terms of Speech Activity Detection error rate (SAD). As result, a best SAD equal to 5.8% and 2.6% is reached respectively in the two considered datasets. In addiction, a significant solidity in terms of microphone positioning is observed in the case of CNN.

Paolo Vecchiotti

Papers

A neural network based algorithm for speaker localization in a multi-room environment

End-to-end Binaural Sound Localisation from the Raw Waveform

Localizing speakers in multiple rooms by using Deep Neural Networks

Deep neural networks for Multi-Room Voice Activity Detection: Advancements and comparative evaluation

End-to-end Binaural Sound Localisation from the Raw Waveform