Electrical Engineering And Systems Science

This report describes the speaker diarization system developed by the ABSP Laboratory team for the third DIHARD speech diarization challenge. Our primary contribution is to develop acoustic domain identification (ADI) system for speaker diarization. We investigate speaker embeddings based ADI system. We apply a domain-dependent threshold for agglomerative hierarchical clustering. Besides, we optimize the parameters for PCA-based dimensionality reduction in a domain-dependent way. Our method of integrating domain-based processing schemes in the baseline system of the challenge achieved a relative improvement of 9.63% and 10.64% in DER for core and full conditions, respectively, for Track 1 of the DIHARD III evaluation set.

Speech separation algorithms are often used to separate the target speech from other interfering sources. However, purely neural network based speech separation systems often cause nonlinear distortion that is harmful for automatic speech recognition (ASR) systems. The conventional mask-based minimum variance distortionless response (MVDR) beamformer can be used to minimize the distortion, but comes with high level of residual noise. Furthermore, the matrix operations (e.g., matrix inversion) involved in the conventional MVDR solution are sometimes numerically unstable when jointly trained with neural networks. In this paper, we propose a novel all deep learning MVDR framework, where the matrix inversion and eigenvalue decomposition are replaced by two recurrent neural networks (RNNs), to resolve both issues at the same time. The proposed method can greatly reduce the residual noise while keeping the target speech undistorted by leveraging on the RNN-predicted frame-wise beamforming weights. The system is evaluated on a Mandarin audio-visual corpus and compared against several state-of-the-art (SOTA) speech separation systems. Experimental results demonstrate the superiority of the proposed method across several objective metrics and ASR accuracy.

Audio-visual speech enhancement system is regarded to be one of promising solutions for isolating and enhancing speech of desired speaker. Conventional methods focus on predicting clean speech spectrum via a naive convolution neural network based encoder-decoder architecture, and these methods a) not adequate to use data fully and effectively, b) cannot process features selectively. The proposed model addresses these drawbacks, by a) applying a model that fuses audio and visual features layer by layer in encoding phase, and that feeds fused audio-visual features to each corresponding decoder layer, and more importantly, b) introducing soft threshold attention into the model to select the informative modality softly. This paper proposes attentional audio-visual multi-layer feature fusion model, in which soft threshold attention unit are applied on feature mapping at every layer of decoder. The proposed model demonstrates the superior performance of the network against the state-of-the-art models.

Speech is converted to digital signals using speech coding for efficient transmission. However, this often lowers the quality and bandwidth of speech. This paper explores the application of convolutional neural networks for Artificial Bandwidth Expansion (ABE) and speech enhancement on coded speech, particularly Adaptive Multi-Rate (AMR) used in 2G cellular phone calls. In this paper, we introduce AMRConvNet: a convolutional neural network that performs ABE and speech enhancement on speech encoded with AMR. The model operates directly on the time-domain for both input and output speech but optimizes using combined time-domain reconstruction loss and frequency-domain perceptual loss. AMRConvNet resulted in an average improvement of 0.425 Mean Opinion Score - Listening Quality Objective (MOS-LQO) points for AMR bitrate of 4.75k, and 0.073 MOS-LQO points for AMR bitrate of 12.2k. AMRConvNet also showed robustness in AMR bitrate inputs. Finally, an ablation test showed that our combined time-domain and frequency-domain loss leads to slightly higher MOS-LQO and faster training convergence than using either loss alone.

The ASVspoof initiative was conceived to spearhead research in anti-spoofing for automatic speaker verification (ASV). This paper describes the third in a series of bi-annual challenges: ASVspoof 2019. With the challenge database and protocols being described elsewhere, the focus of this paper is on results and the top performing single and ensemble system submissions from 62 teams, all of which out-perform the two baseline systems, often by a substantial margin. Deeper analyses shows that performance is dominated by specific conditions involving either specific spoofing attacks or specific acoustic environments. While fusion is shown to be particularly effective for the logical access scenario involving speech synthesis and voice conversion attacks, participants largely struggled to apply fusion successfully for the physical access scenario involving simulated replay attacks. This is likely the result of a lack of system complementarity, while oracle fusion experiments show clear potential to improve performance. Furthermore, while results for simulated data are promising, experiments with real replay data show a substantial gap, most likely due to the presence of additive noise in the latter. This finding, among others, leads to a number of ideas for further research and directions for future editions of the ASVspoof challenge.

End-to-end neural network models (E2E) have shown significant performance benefits on different INTERSPEECH ComParE tasks. Prior work has applied either a single instance of an E2E model for a task or the same E2E architecture for different tasks. However, applying a single model is unstable or using the same architecture under-utilizes task-specific information. On ComParE 2020 tasks, we investigate applying an ensemble of E2E models for robust performance and developing task-specific modifications for each task. ComParE 2020 introduces three sub-challenges: the breathing sub-challenge to predict the output of a respiratory belt worn by a patient while speaking, the elderly sub-challenge to estimate the elderly speaker's arousal and valence levels and the mask sub-challenge to classify if the speaker is wearing a mask or not. On each of these tasks, an ensemble outperforms the single E2E model. On the breathing sub-challenge, we study the impact of multi-loss strategies on task performance. On the elderly sub-challenge, predicting the valence and arousal levels prompts us to investigate multi-task training and implement data sampling strategies to handle class imbalance. On the mask sub-challenge, using an E2E system without feature engineering is competitive to feature-engineered baselines and provides substantial gains when combined with feature-engineered baselines.

Independent Vector Analysis (IVA) is an effective approach for Blind Source Separation (BSS) of convolutive mixtures of audio signals. As a practical realization of an IVA-based BSS algorithm, the so-called AuxIVA update rules based on the Majorize-Minimize (MM) principle have been proposed which allow for fast and computationally efficient optimization of the IVA cost function. For many real-time applications, however, update rules for IVA exhibiting even faster convergence are highly desirable. To this end, we investigate techniques which accelerate the convergence of the AuxIVA update rules without extra computational cost. The efficacy of the proposed methods is verified in experiments representing real-world acoustic scenarios.

Small footprint embedded devices require keyword spotters (KWS) with small model size and detection latency for enabling voice assistants. Such a keyword is often referred to as \textit{wake word} as it is used to wake up voice assistant enabled devices. Together with wake word detection, accurate estimation of wake word endpoints (start and end) is an important task of KWS. In this paper, we propose two new methods for detecting the endpoints of wake words in neural KWS that use single-stage word-level neural networks. Our results show that the new techniques give superior accuracy for detecting wake words' endpoints of up to 50 msec standard error versus human annotations, on par with the conventional Acoustic Model plus HMM forced alignment. To our knowledge, this is the first study of wake word endpoints detection methods for single-stage neural KWS.

This paper proposes a novel acoustic word embedding called Acoustic Neighbor Embeddings where speech or text of arbitrary length are mapped to a vector space of fixed, reduced dimensions by adapting stochastic neighbor embedding (SNE) to sequential inputs. The Euclidean distance between coordinates in the embedding space reflects the phonetic confusability between their corresponding sequences. Two encoder neural networks are trained: an acoustic encoder that accepts speech signals in the form of frame-wise subword posterior probabilities obtained from an acoustic model and a text encoder that accepts text in the form of subword transcriptions. Compared to a triplet loss criterion, the proposed method is shown to have more effective gradients for neural network training. Experimentally, it also gives more accurate results with low-dimensional embeddings when the two encoder networks are used in tandem in a word (name) recognition task, and when the text encoder network is used standalone in an approximate phonetic matching task. In particular, in an isolated name recognition task depending solely on Euclidean nearest-neighbor search between the proposed embedding vectors, the recognition accuracy is identical to that of conventional finite state transducer(FST)-based decoding using test data with up to 1 million names in the vocabulary and 40 dimensions in the embeddings.

Face masks muffle speech and make communication more difficult, especially for people with hearing loss. This study examines the acoustic attenuation caused by different face masks, including medical, cloth, and transparent masks, using a head-shaped loudspeaker and a live human talker. The results suggest that all masks attenuate frequencies above 1 kHz, that attenuation is greatest in front of the talker, and that there is substantial variation between mask types, especially cloth masks with different materials and weaves. Transparent masks have poor acoustic performance compared to both medical and cloth masks. Most masks have little effect on lapel microphones, suggesting that existing sound reinforcement and assistive listening systems may be effective for verbal communication with masks.