Electrical Engineering And Systems Science

End-to-end (E2E) automatic speech recognition (ASR) systems lack the distinct language model (LM) component that characterizes traditional speech systems. While this simplifies the model architecture, it complicates the task of incorporating text-only data into training, which is important to the recognition of tail words that do not occur often in audio-text pairs. While shallow fusion has been proposed as a method for incorporating a pre-trained LM into an E2E model at inference time, it has not yet been explored for very large text corpora, and it has been shown to be very sensitive to hyperparameter settings in the beam search. In this work, we apply shallow fusion to incorporate a very large text corpus into a state-of-the-art E2EASR model. We explore the impact of model size and show that intelligent pruning of the training set can be more effective than increasing the parameter count. Additionally, we show that incorporating the LM in minimum word error rate (MWER) fine tuning makes shallow fusion far less dependent on optimal hyperparameter settings, reducing the difficulty of that tuning problem.

Information on speaker characteristics can be useful as side information in improving speaker recognition accuracy. However, such information is often private. This paper investigates how privacy-preserving learning can improve a speaker verification system, by enabling the use of privacy-sensitive speaker data to train an auxiliary classification model that predicts vocal characteristics of speakers. In particular, this paper explores the utility achieved by approaches which combine different federated learning and differential privacy mechanisms. These approaches make it possible to train a central model while protecting user privacy, with users' data remaining on their devices. Furthermore, they make learning on a large population of speakers possible, ensuring good coverage of speaker characteristics when training a model. The auxiliary model described here uses features extracted from phrases which trigger a speaker verification system. From these features, the model predicts speaker characteristic labels considered useful as side information. The knowledge of the auxiliary model is distilled into a speaker verification system using multi-task learning, with the side information labels predicted by this auxiliary model being the additional task. This approach results in a 6% relative improvement in equal error rate over a baseline system.

In noisy conditions, knowing speech contents facilitates listeners to more effectively suppress background noise components and to retrieve pure speech signals. Previous studies have also confirmed the benefits of incorporating phonetic information in a speech enhancement (SE) system to achieve better denoising performance. To obtain the phonetic information, we usually prepare a phoneme-based acoustic model, which is trained using speech waveforms and phoneme labels. Despite performing well in normal noisy conditions, when operating in very noisy conditions, however, the recognized phonemes may be erroneous and thus misguide the SE process. To overcome the limitation, this study proposes to incorporate the broad phonetic class (BPC) information into the SE process. We have investigated three criteria to build the BPC, including two knowledge-based criteria: place and manner of articulatory and one data-driven criterion. Moreover, the recognition accuracies of BPCs are much higher than that of phonemes, thus providing more accurate phonetic information to guide the SE process under very noisy conditions. Experimental results demonstrate that the proposed SE with the BPC information framework can achieve notable performance improvements over the baseline system and an SE system using monophonic information in terms of both speech quality intelligibility on the TIMIT dataset.

Modern approaches to text to speech require the entire input character sequence to be processed before any audio is synthesised. This latency limits the suitability of such models for time-sensitive tasks like simultaneous interpretation. Interleaving the action of reading a character with that of synthesising audio reduces this latency. However, the order of this sequence of interleaved actions varies across sentences, which raises the question of how the actions should be chosen. We propose a reinforcement learning based framework to train an agent to make this decision. We compare our performance against that of deterministic, rule-based systems. Our results demonstrate that our agent successfully balances the trade-off between the latency of audio generation and the quality of synthesised audio. More broadly, we show that neural sequence-to-sequence models can be adapted to run in an incremental manner.

This paper studies the density priors for independent vector analysis (IVA) with convolutive speech mixture separation as the exemplary application. Most existing source priors for IVA are too simplified to capture the fine structures of speeches. Here, we first time show that it is possible to efficiently estimate the derivative of speech density with universal approximators like deep neural networks (DNN) by optimizing certain proxy separation related performance indices. Experimental results suggest that the resultant neural network density priors consistently outperform previous ones in convergence speed for online implementation and signal-to-interference ratio (SIR) for batch implementation.

We propose an approach of graph convolutional networks for robust infant cry classification. We construct non-fully connected graphs based on the similarities among the relevant nodes in both supervised and semi-supervised node classification with convolutional neural networks to consider the short-term and long-term effects of infant cry signals related to inner-class and inter-class messages. The approach captures the diversity of variations within infant cries, especially for limited training samples. The effectiveness of this approach is evaluated on Baby Chillanto Database and Baby2020 database. With as limited as 20% of labeled training data, our model outperforms that of CNN model with 80% labeled training data and the accuracy stably improves as the number of labeled training samples increases. The best results give significant improvements of 7.36% and 3.59% compared with the results of the CNN models on Baby Chillanto database and Baby2020 database respectively.

Attentive listening in a multispeaker environment such as a cocktail party requires suppression of the interfering speakers and the noise around. People with normal hearing perform remarkably well in such situations. Analysis of the cortical signals using electroencephalography (EEG) has revealed that the EEG signals track the envelope of the attended speech stronger than that of the interfering speech. This has enabled the development of algorithms that can decode the selective attention of a listener in controlled experimental settings. However, often these algorithms require longer trial duration and computationally expensive calibration to obtain a reliable inference of attention. In this paper, we present a novel framework to decode the attention of a listener within trial durations of the order of two seconds. It comprises of three modules: 1) Dynamic estimation of the temporal response functions (TRF) in every trial using a sequential linear minimum mean squared error (LMMSE) estimator, 2) Extract the N1-P2 peak of the estimated TRF that serves as a marker related to the attentional state and 3) Obtain a probabilistic measure of the attentional state using a support vector machine followed by a logistic regression. The efficacy of the proposed decoding framework was evaluated using EEG data collected from 27 subjects. The total number of electrodes required to infer the attention was four: One for the signal estimation, one for the noise estimation and the other two being the reference and the ground electrodes. Our results make further progress towards the realization of neuro-steered hearing aids.

In this paper, we seek solutions for reducing the computation complexity of transformer-based models for speech representation learning. We evaluate 10 attention algorithms; then, we pre-train the transformer-based model with those attention algorithms in a self-supervised fashion and treat them as feature extractors on downstream tasks, including phoneme classification and speaker classification. With the assistance of t-SNE, PCA and some observation, the attention weights in self-supervised audio transformers can be categorized into four general cases. Based on these cases and some analyses, we are able to use a specific set of attention weights to initialize the model. Our approach shows comparable performance to the typical self-attention yet requires 20% less time in both training and inference.

Power spectral density (PSD) estimates of various microphone signal components are essential to many speech enhancement procedures. As speech is highly non-nonstationary, performance improvements may be gained by maintaining time-variations in PSD estimates. In this paper, we propose an instantaneous PSD estimation approach based on generalized principal components. Similarly to other eigenspace-based PSD estimation approaches, we rely on recursive averaging in order to obtain a microphone signal correlation matrix estimate to be decomposed. However, instead of estimating the PSDs directly from the temporally smooth generalized eigenvalues of this matrix, yielding temporally smooth PSD estimates, we propose to estimate the PSDs from newly defined instantaneous generalized eigenvalues, yielding instantaneous PSD estimates. The instantaneous generalized eigenvalues are defined from the generalized principal components, i.e. a generalized eigenvector-based transform of the microphone signals. We further show that the smooth generalized eigenvalues can be understood as a recursive average of the instantaneous generalized eigenvalues. Simulation results comparing the multi-channel Wiener filter (MWF) with smooth and instantaneous PSD estimates indicate better speech enhancement performance for the latter. A MATLAB implementation is available online.

The task for speaker verification (SV) is to decide an utterance is spoken by a target or imposter speaker. In most SV studies, a log-likelihood ratio (L_LLR) score is estimated based on a generative probability model on speaker features, and compared with a threshold for decision making. However, the generative model usually focuses on feature distributions and does not have the discriminative feature selection ability, which is easy to be distracted by nuisance features. The SV, as a hypothesis test, could be formulated as a binary classification task where a neural network (NN) based discriminative learning could be applied. Through discriminative learning, the nuisance features could be removed with the help of label supervision. However, the discriminative learning pays more attention to classification boundaries which is prone to overfitting to training data and yielding poor generalization on testing data. In this paper, we propose a hybrid learning framework, i.e., integrating a joint Bayesian (JB) generative model into a neural discriminative learning framework for SV. A Siamese NN is built with dense layers to approximate the mapping functions used in the SV pipeline with the JB model, and the L-LLR score estimated based on the JB model is connected to the distance metric in a pair-wised discriminative learning. By initializing the Siamese NN with the parameters learned from the JB model, we further train the model parameters with the pair-wised samples as a binary discrimination task. Moreover, direct evaluation metric in SV, i.e., minimum empirical Bayes risk, is designed and integrated as an objective function in the discriminative learning. We carried out SV experiments on speakers in the wild (SITW) and Voxceleb corpora. Experimental results showed that our proposed model improved the performance with a large margin compared with state-of-the-art models for SV.