Thomas Fang Zheng

Qiuming Zhao, Guangzhi Sun, Chao Zhang et al.

Recent end-to-end automatic speech recognition (ASR) models have become increasingly larger, making them particularly challenging to be deployed on resource-constrained devices. Model quantisation is an effective solution that sometimes causes the word error rate (WER) to increase. In this paper, a novel strategy of personalisation for a quantised model (PQM) is proposed, which combines speaker adaptive training (SAT) with model quantisation to improve the performance of heavily compressed models. Specifically, PQM uses a 4-bit NormalFloat Quantisation (NF4) approach for model quantisation and low-rank adaptation (LoRA) for SAT. Experiments have been performed on the LibriSpeech and the TED-LIUM 3 corpora. Remarkably, with a 7x reduction in model size and 1% additional speaker-specific parameters, 15.1% and 23.3% relative WER reductions were achieved on quantised Whisper and Conformer-based attention-based encoder-decoder ASR models respectively, comparing to the original full precision models.

Turi Abu, Ying Shi, Thomas Fang Zheng et al.

We present a novel Automatic Speech Recognition (ASR) dataset for the Oromo language, a widely spoken language in Ethiopia and neighboring regions. The dataset was collected through a crowd-sourcing initiative, encompassing a diverse range of speakers and phonetic variations. It consists of 100 hours of real-world audio recordings paired with transcriptions, covering read speech in both clean and noisy environments. This dataset addresses the critical need for ASR resources for the Oromo language which is underrepresented. To show its applicability for the ASR task, we conducted experiments using the Conformer model, achieving a Word Error Rate (WER) of 15.32% with hybrid CTC and AED loss and WER of 18.74% with pure CTC loss. Additionally, fine-tuning the Whisper model resulted in a significantly improved WER of 10.82%. These results establish baselines for Oromo ASR, highlighting both the challenges and the potential for improving ASR performance in Oromo. The dataset is publicly available at https://github.com/turinaf/sagalee and we encourage its use for further research and development in Oromo speech processing.

Haoran Sun, Lantian Li, Thomas Fang Zheng et al.

The way that humans encode their emotion into speech signals is complex. For instance, an angry man may increase his pitch and speaking rate, and use impolite words. In this paper, we present a preliminary study on various emotional factors and investigate how each of them impacts modern emotion recognition systems. The key tool of our study is the SpeechFlow model presented recently, by which we are able to decompose speech signals into separate information factors (content, pitch, rhythm). Based on this decomposition, we carefully studied the performance of each information component and their combinations. We conducted the study on three different speech emotion corpora and chose an attention-based convolutional RNN as the emotion classifier. Our results show that rhythm is the most important component for emotional expression. Moreover, the cross-corpus results are very bad (even worse than guess), demonstrating that the present speech emotion recognition model is rather weak. Interestingly, by removing one or several unimportant components, the cross-corpus results can be improved. This demonstrates the potential of the decomposition approach towards a generalizable emotion recognition.

Wei Liu, Meng Sun, Xiongwei Zhang et al.

The choice of an optimal time-frequency resolution is usually a difficult but important step in tasks involving speech signal classification, e.g., speech anti-spoofing. The variations of the performance with different choices of timefrequency resolutions can be as large as those with different model architectures, which makes it difficult to judge what the improvement actually comes from when a new network architecture is invented and introduced as the classifier. In this paper, we propose a multi-resolution front-end for feature extraction in an end-to-end classification framework. Optimal weighted combinations of multiple time-frequency resolutions will be learned automatically given the objective of a classification task. Features extracted with different time-frequency resolutions are weighted and concatenated as inputs to the successive networks, where the weights are predicted by a learnable neural network inspired by the weighting block in squeeze-and-excitation networks (SENet). Furthermore, the refinement of the chosen timefrequency resolutions is investigated by pruning the ones with relatively low importance, which reduces the complexity and size of the model. The proposed method is evaluated on the tasks of speech anti-spoofing in ASVSpoof 2019 and its superiority has been justified by comparing with similar baselines.

Weiyi Zhang, Shuning Zhao, Le Liu et al.

In authentication scenarios, applications of practical speaker verification systems usually require a person to read a dynamic authentication text. Previous studies played an audio adversarial example as a digital signal to perform physical attacks, which would be easily rejected by audio replay detection modules. This work shows that by playing our crafted adversarial perturbation as a separate source when the adversary is speaking, the practical speaker verification system will misjudge the adversary as a target speaker. A two-step algorithm is proposed to optimize the universal adversarial perturbation to be text-independent and has little effect on the authentication text recognition. We also estimated room impulse response (RIR) in the algorithm which allowed the perturbation to be effective after being played over the air. In the physical experiment, we achieved targeted attacks with success rate of 100%, while the word error rate (WER) on speech recognition was only increased by 3.55%. And recorded audios could pass replay detection for the live person speaking.

Lantian Li, Ruiqi Liu, Jiawen Kang et al.

Research on speaker recognition is extending to address the vulnerability in the wild conditions, among which genre mismatch is perhaps the most challenging, for instance, enrollment with reading speech while testing with conversational or singing audio. This mismatch leads to complex and composite inter-session variations, both intrinsic (i.e., speaking style, physiological status) and extrinsic (i.e., recording device, background noise). Unfortunately, the few existing multi-genre corpora are not only limited in size but are also recorded under controlled conditions, which cannot support conclusive research on the multi-genre problem. In this work, we firstly publish CN-Celeb, a large-scale multi-genre corpus that includes in-the-wild speech utterances of 3,000 speakers in 11 different genres. Secondly, using this dataset, we conduct a comprehensive study on the multi-genre phenomenon, in particular the impact of the multi-genre challenge on speaker recognition and the performance gain when the new dataset is used to conduct multi-genre training.

Lantian Li, Yang Zhang, Jiawen Kang et al.

Domain mismatch often occurs in real applications and causes serious performance reduction on speaker verification systems. The common wisdom is to collect cross-domain data and train a multi-domain PLDA model, with the hope to learn a domain-independent speaker subspace. In this paper, we firstly present an empirical study to show that simply adding cross-domain data does not help performance in conditions with enrollment-test mismatch. Careful analysis shows that this striking result is caused by the incoherent statistics between the enrollment and test conditions. Based on this analysis, we present a decoupled scoring approach that can maximally squeeze the value of cross-domain labels and obtain optimal verification scores when the enrollment and test are mismatched. When the statistics are coherent, the new formulation falls back to the conventional PLDA. Experimental results on cross-channel test show that the proposed approach is highly effective and is a principle solution to domain mismatch.

Haoran Sun, Lantian Li, Yunqi Cai et al.

Various information factors are blended in speech signals, which forms the primary difficulty for most speech information processing tasks. An intuitive idea is to factorize speech signal into individual information factors (e.g., phonetic content and speaker trait), though it turns out to be highly challenging. This paper presents a speech factorization approach based on a novel factorial discriminative normalization flow model (factorial DNF). Experiments conducted on a two-factor case that involves phonetic content and speaker trait demonstrates that the proposed factorial DNF has powerful capability to factorize speech signals and outperforms several comparative models in terms of information representation and manipulation.

Linlin Zheng, Jiakang Li, Meng Sun et al.

The technique of transforming voices in order to hide the real identity of a speaker is called voice disguise, among which automatic voice disguise (AVD) by modifying the spectral and temporal characteristics of voices with miscellaneous algorithms are easily conducted with softwares accessible to the public. AVD has posed great threat to both human listening and automatic speaker verification (ASV). In this paper, we have found that ASV is not only a victim of AVD but could be a tool to beat some simple types of AVD. Firstly, three types of AVD, pitch scaling, vocal tract length normalization (VTLN) and voice conversion (VC), are introduced as representative methods. State-of-the-art ASV methods are subsequently utilized to objectively evaluate the impact of AVD on ASV by equal error rates (EER). Moreover, an approach to restore disguised voice to its original version is proposed by minimizing a function of ASV scores w.r.t. restoration parameters. Experiments are then conducted on disguised voices from Voxceleb, a dataset recorded in real-world noisy scenario. The results have shown that, for the voice disguise by pitch scaling, the proposed approach obtains an EER around 7% comparing to the 30% EER of a recently proposed baseline using the ratio of fundamental frequencies. The proposed approach generalizes well to restore the disguise with nonlinear frequency warping in VTLN by reducing its EER from 34.3% to 18.5%. However, it is difficult to restore the source speakers in VC by our approach, where more complex forms of restoration functions or other paralinguistic cues might be necessary to restore the nonlinear transform in VC. Finally, contrastive visualization on ASV features with and without restoration illustrate the role of the proposed approach in an intuitive way.

Lantian Li, Dong Wang, Yixiang Chen et al.

Various informative factors mixed in speech signals, leading to great difficulty when decoding any of the factors. An intuitive idea is to factorize each speech frame into individual informative factors, though it turns out to be highly difficult. Recently, we found that speaker traits, which were assumed to be long-term distributional properties, are actually short-time patterns, and can be learned by a carefully designed deep neural network (DNN). This discovery motivated a cascade deep factorization (CDF) framework that will be presented in this paper. The proposed framework infers speech factors in a sequential way, where factors previously inferred are used as conditional variables when inferring other factors. We will show that this approach can effectively factorize speech signals, and using these factors, the original speech spectrum can be recovered with a high accuracy. This factorization and reconstruction approach provides potential values for many speech processing tasks, e.g., speaker recognition and emotion recognition, as will be demonstrated in the paper.

Lantian Li, Zhiyuan Tang, Dong Wang et al.

In recent studies, it has shown that speaker patterns can be learned from very short speech segments (e.g., 0.3 seconds) by a carefully designed convolutional & time-delay deep neural network (CT-DNN) model. By enforcing the model to discriminate the speakers in the training data, frame-level speaker features can be derived from the last hidden layer. In spite of its good performance, a potential problem of the present model is that it involves a parametric classifier, i.e., the last affine layer, which may consume some discriminative knowledge, thus leading to `information leak' for the feature learning. This paper presents a full-info training approach that discards the parametric classifier and enforces all the discriminative knowledge learned by the feature net. Our experiments on the Fisher database demonstrate that this new training scheme can produce more coherent features, leading to consistent and notable performance improvement on the speaker verification task.

Aodong Li, Shiyue Zhang, Dong Wang et al.

Neural machine translation (NMT) has recently achieved impressive results. A potential problem of the existing NMT algorithm, however, is that the decoding is conducted from left to right, without considering the right context. This paper proposes an two-stage approach to solve the problem. In the first stage, a conventional attention-based NMT system is used to produce a draft translation, and in the second stage, a novel double-attention NMT system is used to refine the translation, by looking at the original input as well as the draft translation. This drafting-and-refinement can obtain the right-context information from the draft, hence producing more consistent translations. We evaluated this approach using two Chinese-English translation tasks, one with 44k pairs and 1M pairs respectively. The experiments showed that our approach achieved positive improvements over the conventional NMT system: the improvements are 2.4 and 0.9 BLEU points on the small-scale and large-scale tasks, respectively.

Dong Wang, Lantian Li, Zhiyuan Tang et al.

End-to-end learning treats the entire system as a whole adaptable black box, which, if sufficient data are available, may learn a system that works very well for the target task. This principle has recently been applied to several prototype research on speaker verification (SV), where the feature learning and classifier are learned together with an objective function that is consistent with the evaluation metric. An opposite approach to end-to-end is feature learning, which firstly trains a feature learning model, and then constructs a back-end classifier separately to perform SV. Recently, both approaches achieved significant performance gains on SV, mainly attributed to the smart utilization of deep neural networks. However, the two approaches have not been carefully compared, and their respective advantages have not been well discussed. In this paper, we compare the end-to-end and feature learning approaches on a text-independent SV task. Our experiments on a dataset sampled from the Fisher database and involving 5,000 speakers demonstrated that the feature learning approach outperformed the end-to-end approach. This is a strong support for the feature learning approach, at least with data and computation resources similar to ours.

Lantian Li, Dong Wang, Askar Rozi et al.

Existing speaker verification (SV) systems often suffer from performance degradation if there is any language mismatch between model training, speaker enrollment, and test. A major cause of this degradation is that most existing SV methods rely on a probabilistic model to infer the speaker factor, so any significant change on the distribution of the speech signal will impact the inference. Recently, we proposed a deep learning model that can learn how to extract the speaker factor by a deep neural network (DNN). By this feature learning, an SV system can be constructed with a very simple back-end model. In this paper, we investigate the robustness of the feature-based SV system in situations with language mismatch. Our experiments were conducted on a complex cross-lingual scenario, where the model training was in English, and the enrollment and test were in Chinese or Uyghur. The experiments demonstrated that the feature-based system outperformed the i-vector system with a large margin, particularly with language mismatch between enrollment and test.

Lantian Li, Yixiang Chen, Dong Wang et al.

For practical automatic speaker verification (ASV) systems, replay attack poses a true risk. By replaying a pre-recorded speech signal of the genuine speaker, ASV systems tend to be easily fooled. An effective replay detection method is therefore highly desirable. In this study, we investigate a major difficulty in replay detection: the over-fitting problem caused by variability factors in speech signal. An F-ratio probing tool is proposed and three variability factors are investigated using this tool: speaker identity, speech content and playback & recording device. The analysis shows that device is the most influential factor that contributes the highest over-fitting risk. A frequency warping approach is studied to alleviate the over-fitting problem, as verified on the ASV-spoof 2017 database.

Lantian Li, Renyu Wang, Gang Wang et al.

Decision making is an important component in a speaker verification system. For the conventional GMM-UBM architecture, the decision is usually conducted based on the log likelihood ratio of the test utterance against the GMM of the claimed speaker and the UBM. This single-score decision is simple but tends to be sensitive to the complex variations in speech signals (e.g. text content, channel, speaking style, etc.). In this paper, we propose a decision making approach based on multiple scores derived from a set of cohort GMMs (cohort scores). Importantly, these cohort scores are not simply averaged as in conventional cohort methods; instead, we employ a powerful discriminative model as the decision maker. Experimental results show that the proposed method delivers substantial performance improvement over the baseline system, especially when a deep neural network (DNN) is used as the decision maker, and the DNN input involves some statistical features derived from the cohort scores.

Lantian Li, Dong Wang, Xiaodong Zhang et al.

For text-independent short-utterance speaker recognition (SUSR), the performance often degrades dramatically. This paper presents a combination approach to the SUSR tasks with two phonetic-aware systems: one is the DNN-based i-vector system and the other is our recently proposed subregion-based GMM-UBM system. The former employs phone posteriors to construct an i-vector model in which the shared statistics offers stronger robustness against limited test data, while the latter establishes a phone-dependent GMM-UBM system which represents speaker characteristics with more details. A score-level fusion is implemented to integrate the respective advantages from the two systems. Experimental results show that for the text-independent SUSR task, both the DNN-based i-vector system and the subregion-based GMM-UBM system outperform their respective baselines, and the score-level system combination delivers performance improvement.

Dong Wang, Thomas Fang Zheng

Transfer learning is a vital technique that generalizes models trained for one setting or task to other settings or tasks. For example in speech recognition, an acoustic model trained for one language can be used to recognize speech in another language, with little or no re-training data. Transfer learning is closely related to multi-task learning (cross-lingual vs. multilingual), and is traditionally studied in the name of `model adaptation'. Recent advance in deep learning shows that transfer learning becomes much easier and more effective with high-level abstract features learned by deep models, and the `transfer' can be conducted not only between data distributions and data types, but also between model structures (e.g., shallow nets and deep nets) or even model types (e.g., Bayesian models and neural models). This review paper summarizes some recent prominent research towards this direction, particularly for speech and language processing. We also report some results from our group and highlight the potential of this very interesting research field.

Lantian Li, Dong Wang, Chao Xing et al.

Probabilistic linear discriminant analysis (PLDA) is a popular normalization approach for the i-vector model, and has delivered state-of-the-art performance in speaker recognition. A potential problem of the PLDA model, however, is that it essentially assumes Gaussian distributions over speaker vectors, which is not always true in practice. Additionally, the objective function is not directly related to the goal of the task, e.g., discriminating true speakers and imposters. In this paper, we propose a max-margin metric learning approach to solve the problems. It learns a linear transform with a criterion that the margin between target and imposter trials are maximized. Experiments conducted on the SRE08 core test show that compared to PLDA, the new approach can obtain comparable or even better performance, though the scoring is simply a cosine computation.

Lantian Li, Dong Wang, Chao Xing et al.

The popular i-vector model represents speakers as low-dimensional continuous vectors (i-vectors), and hence it is a way of continuous speaker embedding. In this paper, we investigate binary speaker embedding, which transforms i-vectors to binary vectors (codes) by a hash function. We start from locality sensitive hashing (LSH), a simple binarization approach where binary codes are derived from a set of random hash functions. A potential problem of LSH is that the randomly sampled hash functions might be suboptimal. We therefore propose an improved Hamming distance learning approach, where the hash function is learned by a variable-sized block training that projects each dimension of the original i-vectors to variable-sized binary codes independently. Our experiments show that binary speaker embedding can deliver competitive or even better results on both speaker verification and identification tasks, while the memory usage and the computation cost are significantly reduced.

Lantian Li, Dong Wang, Zhiyong Zhang et al.

Recent research shows that deep neural networks (DNNs) can be used to extract deep speaker vectors (d-vectors) that preserve speaker characteristics and can be used in speaker verification. This new method has been tested on text-dependent speaker verification tasks, and improvement was reported when combined with the conventional i-vector method. This paper extends the d-vector approach to semi text-independent speaker verification tasks, i.e., the text of the speech is in a limited set of short phrases. We explore various settings of the DNN structure used for d-vector extraction, and present a phone-dependent training which employs the posterior features obtained from an ASR system. The experimental results show that it is possible to apply d-vectors on semi text-independent speaker recognition, and the phone-dependent training improves system performance.

Miao Fan, Qiang Zhou, Thomas Fang Zheng

Entity linking is an indispensable operation of populating knowledge repositories for information extraction. It studies on aligning a textual entity mention to its corresponding disambiguated entry in a knowledge repository. In this paper, we propose a new paradigm named distantly supervised entity linking (DSEL), in the sense that the disambiguated entities that belong to a huge knowledge repository (Freebase) are automatically aligned to the corresponding descriptive webpages (Wiki pages). In this way, a large scale of weakly labeled data can be generated without manual annotation and fed to a classifier for linking more newly discovered entities. Compared with traditional paradigms based on solo knowledge base, DSEL benefits more via jointly leveraging the respective advantages of Freebase and Wikipedia. Specifically, the proposed paradigm facilitates bridging the disambiguated labels (Freebase) of entities and their textual descriptions (Wikipedia) for Web-scale entities. Experiments conducted on a dataset of 140,000 items and 60,000 features achieve a baseline F1-measure of 0.517. Furthermore, we analyze the feature performance and improve the F1-measure to 0.545.

Miao Fan, Qiang Zhou, Andrew Abel et al.

This paper contributes a novel embedding model which measures the probability of each belief $\langle h,r,t,m\rangle$ in a large-scale knowledge repository via simultaneously learning distributed representations for entities ($h$ and $t$), relations ($r$), and the words in relation mentions ($m$). It facilitates knowledge completion by means of simple vector operations to discover new beliefs. Given an imperfect belief, we can not only infer the missing entities, predict the unknown relations, but also tell the plausibility of the belief, just leveraging the learnt embeddings of remaining evidences. To demonstrate the scalability and the effectiveness of our model, we conduct experiments on several large-scale repositories which contain millions of beliefs from WordNet, Freebase and NELL, and compare it with other cutting-edge approaches via competing the performances assessed by the tasks of entity inference, relation prediction and triplet classification with respective metrics. Extensive experimental results show that the proposed model outperforms the state-of-the-arts with significant improvements.

Miao Fan, Qiang Zhou, Thomas Fang Zheng et al.

Traditional way of storing facts in triplets ({\it head\_entity, relation, tail\_entity}), abbreviated as ({\it h, r, t}), makes the knowledge intuitively displayed and easily acquired by mankind, but hardly computed or even reasoned by AI machines. Inspired by the success in applying {\it Distributed Representations} to AI-related fields, recent studies expect to represent each entity and relation with a unique low-dimensional embedding, which is different from the symbolic and atomic framework of displaying knowledge in triplets. In this way, the knowledge computing and reasoning can be essentially facilitated by means of a simple {\it vector calculation}, i.e. ${\bf h} + {\bf r} \approx {\bf t}$. We thus contribute an effective model to learn better embeddings satisfying the formula by pulling the positive tail entities ${\bf t^{+}}$ to get together and close to {\bf h} + {\bf r} ({\it Nearest Neighbor}), and simultaneously pushing the negatives ${\bf t^{-}}$ away from the positives ${\bf t^{+}}$ via keeping a {\it Large Margin}. We also design a corresponding learning algorithm to efficiently find the optimal solution based on {\it Stochastic Gradient Descent} in iterative fashion. Quantitative experiments illustrate that our approach can achieve the state-of-the-art performance, compared with several latest methods on some benchmark datasets for two classical applications, i.e. {\it Link prediction} and {\it Triplet classification}. Moreover, we analyze the parameter complexities among all the evaluated models, and analytical results indicate that our model needs fewer computational resources on outperforming the other methods.

Miao Fan, Qiang Zhou, Thomas Fang Zheng

This paper considers the problem of knowledge inference on large-scale imperfect repositories with incomplete coverage by means of embedding entities and relations at the first attempt. We propose IIKE (Imperfect and Incomplete Knowledge Embedding), a probabilistic model which measures the probability of each belief, i.e. $\langle h,r,t\rangle$, in large-scale knowledge bases such as NELL and Freebase, and our objective is to learn a better low-dimensional vector representation for each entity ($h$ and $t$) and relation ($r$) in the process of minimizing the loss of fitting the corresponding confidence given by machine learning (NELL) or crowdsouring (Freebase), so that we can use $||{\bf h} + {\bf r} - {\bf t}||$ to assess the plausibility of a belief when conducting inference. We use subsets of those inexact knowledge bases to train our model and test the performances of link prediction and triplet classification on ground truth beliefs, respectively. The results of extensive experiments show that IIKE achieves significant improvement compared with the baseline and state-of-the-art approaches.

Miao Fan, Deli Zhao, Qiang Zhou et al.

The essence of distantly supervised relation extraction is that it is an incomplete multi-label classification problem with sparse and noisy features. To tackle the sparsity and noise challenges, we propose solving the classification problem using matrix completion on factorized matrix of minimized rank. We formulate relation classification as completing the unknown labels of testing items (entity pairs) in a sparse matrix that concatenates training and testing textual features with training labels. Our algorithmic framework is based on the assumption that the rank of item-by-feature and item-by-label joint matrix is low. We apply two optimization models to recover the underlying low-rank matrix leveraging the sparsity of feature-label matrix. The matrix completion problem is then solved by the fixed point continuation (FPC) algorithm, which can find the global optimum. Experiments on two widely used datasets with different dimensions of textual features demonstrate that our low-rank matrix completion approach significantly outperforms the baseline and the state-of-the-art methods.