Tan Lee

Yujia Xiao, Shaofei Zhang, Xi Wang et al.

While state-of-the-art Text-to-Speech systems can generate natural speech of very high quality at sentence level, they still meet great challenges in speech generation for paragraph / long-form reading. Such deficiencies are due to i) ignorance of cross-sentence contextual information, and ii) high computation and memory cost for long-form synthesis. To address these issues, this work develops a lightweight yet effective TTS system, ContextSpeech. Specifically, we first design a memory-cached recurrence mechanism to incorporate global text and speech context into sentence encoding. Then we construct hierarchically-structured textual semantics to broaden the scope for global context enhancement. Additionally, we integrate linearized self-attention to improve model efficiency. Experiments show that ContextSpeech significantly improves the voice quality and prosody expressiveness in paragraph reading with competitive model efficiency. Audio samples are available at: https://contextspeech.github.io/demo/

Daxin Tan, Liqun Deng, Nianzu Zheng et al.

This study propose a fully automated system for speech correction and accent reduction. Consider the application scenario that a recorded speech audio contains certain errors, e.g., inappropriate words, mispronunciations, that need to be corrected. The proposed system, named CorrectSpeech, performs the correction in three steps: recognizing the recorded speech and converting it into time-stamped symbol sequence, aligning recognized symbol sequence with target text to determine locations and types of required edit operations, and generating the corrected speech. Experiments show that the quality and naturalness of corrected speech depend on the performance of speech recognition and alignment modules, as well as the granularity level of editing operations. The proposed system is evaluated on two corpora: a manually perturbed version of VCTK and L2-ARCTIC. The results demonstrate that our system is able to correct mispronunciation and reduce accent in speech recordings. Audio samples are available online for demonstration https://daxintan-cuhk.github.io/CorrectSpeech/ .

Xu Yang, Juantao Zhong, Daoyuan Wu et al.

Online exams conducted via video conferencing platforms such as Zoom have become widespread, yet ensuring exam integrity remains challenging due to the difficulty of monitoring multiple video feeds in real time. We present iExam, an online exam proctoring and analysis system that combines lightweight real-time face detection with deep face recognition for postexam analysis. iExam assists invigilators by monitoring student presence during exams and identifies abnormal behaviors, such as face disappearance, face rotation, and identity substitution, from recorded videos. The system addresses three key challenges: (i)efficient real-time video capture and analysis, (ii) automated student identity labeling using enhanced OCR on dynamic Zoom name tags, and (iii) resource-efficient training and inference on standard teacher devices. Extensive experiments show that iExam achieves 90.4% accuracy in real-time face detection and 98.4% accuracy in post-exam recognition with low overhead, demonstrating its practicality and effectiveness for online exam proctoring.

Yujia Xiao, Lei He, Haohan Guo et al.

Existing Existing automatic audio generation methods struggle to generate podcast-like audio programs effectively. The key challenges lie in in-depth content generation, appropriate and expressive voice production. This paper proposed PodAgent, a comprehensive framework for creating audio programs. PodAgent 1) generates informative topic-discussion content by designing a Host-Guest-Writer multi-agent collaboration system, 2) builds a voice pool for suitable voice-role matching and 3) utilizes LLM-enhanced speech synthesis method to generate expressive conversational speech. Given the absence of standardized evaluation criteria for podcast-like audio generation, we developed comprehensive assessment guidelines to effectively evaluate the model's performance. Experimental results demonstrate PodAgent's effectiveness, significantly surpassing direct GPT-4 generation in topic-discussion dialogue content, achieving an 87.4% voice-matching accuracy, and producing more expressive speech through LLM-guided synthesis. Demo page: https://podcast-agent.github.io/demo/. Source code: https://github.com/yujxx/PodAgent.

Yujia Xiao, Liumeng Xue, Lei He et al.

Recently, an increasing number of multimodal (text and audio) benchmarks have emerged, primarily focusing on evaluating models' understanding capability. However, exploration into assessing generative capabilities remains limited, especially for open-ended long-form content generation. Significant challenges lie in no reference standard answer, no unified evaluation metrics and uncontrollable human judgments. In this work, we take podcast-like audio generation as a starting point and propose PodEval, a comprehensive and well-designed open-source evaluation framework. In this framework: 1) We construct a real-world podcast dataset spanning diverse topics, serving as a reference for human-level creative quality. 2) We introduce a multimodal evaluation strategy and decompose the complex task into three dimensions: text, speech and audio, with different evaluation emphasis on "Content" and "Format". 3) For each modality, we design corresponding evaluation methods, involving both objective metrics and subjective listening test. We leverage representative podcast generation systems (including open-source, close-source, and human-made) in our experiments. The results offer in-depth analysis and insights into podcast generation, demonstrating the effectiveness of PodEval in evaluating open-ended long-form audio. This project is open-source to facilitate public use: https://github.com/yujxx/PodEval.

Hei Yi Mak, Tan Lee

The majority of inhabitants in Hong Kong are able to read and write in standard Chinese but use Cantonese as the primary spoken language in daily life. Spoken Cantonese can be transcribed into Chinese characters, which constitute the so-called written Cantonese. Written Cantonese exhibits significant lexical and grammatical differences from standard written Chinese. The rise of written Cantonese is increasingly evident in the cyber world. The growing interaction between Mandarin speakers and Cantonese speakers is leading to a clear demand for automatic translation between Chinese and Cantonese. This paper describes a transformer-based neural machine translation (NMT) system for written-Chinese-to-written-Cantonese translation. Given that parallel text data of Chinese and Cantonese are extremely scarce, a major focus of this study is on the effort of preparing good amount of training data for NMT. In addition to collecting 28K parallel sentences from previous linguistic studies and scattered internet resources, we devise an effective approach to obtaining 72K parallel sentences by automatically extracting pairs of semantically similar sentences from parallel articles on Chinese Wikipedia and Cantonese Wikipedia. We show that leveraging highly similar sentence pairs mined from Wikipedia improves translation performance in all test sets. Our system outperforms Baidu Fanyi's Chinese-to-Cantonese translation on 6 out of 8 test sets in BLEU scores. Translation examples reveal that our system is able to capture important linguistic transformations between standard Chinese and spoken Cantonese.

Wei Liu, Jingyong Hou, Dong Yang et al.

Covering all languages with a multilingual speech recognition model (MASR) is very difficult. Performing language extension on top of an existing MASR is a desirable choice. In this study, the MASR continual learning problem is probabilistically decomposed into language identity prediction (LP) and cross-lingual adaptation (XLA) sub-problems. Based on this, we propose an architecture-based framework for language extension that can fundamentally solve catastrophic forgetting, debudded as PELE. PELE is designed to be parameter-efficient, incrementally incorporating an add-on module to adapt to a new language. Specifically, different parameter-efficient fine-tuning (PEFT) modules and their variants are explored as potential candidates to perform XLA. Experiments are carried out on 5 new languages with a wide range of low-resourced data sizes. The best-performing PEFT candidate can achieve satisfactory performance across all languages and demonstrates superiority in three of five languages over the continual joint learning setting. Notably, PEFT methods focusing on weight parameters or input features are revealed to be limited in performance, showing significantly inferior extension capabilities compared to inserting a lightweight module in between layers such as an Adapter.

Guangyan Zhang, Kaitao Song, Xu Tan et al.

Recently, leveraging BERT pre-training to improve the phoneme encoder in text to speech (TTS) has drawn increasing attention. However, the works apply pre-training with character-based units to enhance the TTS phoneme encoder, which is inconsistent with the TTS fine-tuning that takes phonemes as input. Pre-training only with phonemes as input can alleviate the input mismatch but lack the ability to model rich representations and semantic information due to limited phoneme vocabulary. In this paper, we propose MixedPhoneme BERT, a novel variant of the BERT model that uses mixed phoneme and sup-phoneme representations to enhance the learning capability. Specifically, we merge the adjacent phonemes into sup-phonemes and combine the phoneme sequence and the merged sup-phoneme sequence as the model input, which can enhance the model capacity to learn rich contextual representations. Experiment results demonstrate that our proposed Mixed-Phoneme BERT significantly improves the TTS performance with 0.30 CMOS gain compared with the FastSpeech 2 baseline. The Mixed-Phoneme BERT achieves 3x inference speedup and similar voice quality to the previous TTS pre-trained model PnG BERT

Si-Ioi Ng, Rui-Si Ma, Tan Lee et al.

Human speech production encompasses physiological processes that naturally react to physic stress. Stress caused by physical activity (PA), e.g., running, may lead to significant changes in a person's speech. The major changes are related to the aspects of pitch level, speaking rate, pause pattern, and breathiness. The extent of change depends presumably on physical fitness and well-being of the person, as well as intensity of PA. The general wellness of a person is further related to his/her physical literacy (PL), which refers to a holistic description of engagement in PA. This paper presents the development of a Cantonese speech database that contains audio recordings of speech before and after physical exercises of different intensity levels. The corpus design and data collection process are described. Preliminary results of acoustical analysis are presented to illustrate the impact of PA on pitch level, pitch range, speaking and articulation rate, and time duration of pauses. It is also noted that the effect of PA is correlated to some of the PA and PL measures.

Si-Ioi Ng, Tan Lee

Psychoacoustic studies have shown that locally-time reversed (LTR) speech, i.e., signal samples time-reversed within a short segment, can be accurately recognised by human listeners. This study addresses the question of how well a state-of-the-art automatic speech recognition (ASR) system would perform on LTR speech. The underlying objective is to explore the feasibility of deploying LTR speech in the training of end-to-end (E2E) ASR models, as an attempt to data augmentation for improving the recognition performance. The investigation starts with experiments to understand the effect of LTR speech on general-purpose ASR. LTR speech with reversed segment duration of 5 ms - 50 ms is rendered and evaluated. For ASR training data augmentation with LTR speech, training sets are created by combining natural speech with different partitions of LTR speech. The efficacy of data augmentation is confirmed by ASR results on speech corpora in various languages and speaking styles. ASR on LTR speech with reversed segment duration of 15 ms - 30 ms is found to have lower error rate than with other segment duration. Data augmentation with these LTR speech achieves satisfactory and consistent improvement on ASR performance.

Daxin Tan, Guangyan Zhang, Tan Lee

This study aims at designing an environment-aware text-to-speech (TTS) system that can generate speech to suit specific acoustic environments. It is also motivated by the desire to leverage massive data of speech audio from heterogeneous sources in TTS system development. The key idea is to model the acoustic environment in speech audio as a factor of data variability and incorporate it as a condition in the process of neural network based speech synthesis. Two embedding extractors are trained with two purposely constructed datasets for characterization and disentanglement of speaker and environment factors in speech. A neural network model is trained to generate speech from extracted speaker and environment embeddings. Objective and subjective evaluation results demonstrate that the proposed TTS system is able to effectively disentangle speaker and environment factors and synthesize speech audio that carries designated speaker characteristics and environment attribute. Audio samples are available online for demonstration https://daxintan-cuhk.github.io/Environment-Aware-TTS/ .

Guangyan Zhang, Yichong Leng, Daxin Tan et al.

In the development of neural text-to-speech systems, model pre-training with a large amount of non-target speakers' data is a common approach. However, in terms of ultimately achieved system performance for target speaker(s), the actual benefits of model pre-training are uncertain and unstable, depending very much on the quantity and text content of training data. This study aims to understand better why and how model pre-training can positively contribute to TTS system performance. It is postulated that the pre-training process plays a critical role in learning text-related variation in speech, while further training with the target speaker's data aims to capture the speaker-related variation. Different test sets are created with varying degrees of similarity to target speaker data in terms of text content. Experiments show that leveraging a speaker-independent TTS trained on speech data with diverse text content can improve the target speaker TTS on domain-mismatched text. We also attempt to reduce the amount of pre-training data for a new text domain and improve the data and computational efficiency. It is found that the TTS system could achieve comparable performance when the pre-training data is reduced to 1/8 of its original size.

Jingyu Li, Si-Ioi Ng, Tan Lee

The paper presents a novel approach to refining similarity scores between input utterances for robust speaker verification. Given the embeddings from a pair of input utterances, a graph model is designed to incorporate additional information from a group of embeddings representing the so-called auxiliary speakers. The relations between the input utterances and the auxiliary speakers are represented by the edges and vertices in the graph. The similarity scores are refined by iteratively updating the values of the graph's vertices using an algorithm similar to the random walk algorithm on graphs. Through this updating process, the information of auxiliary speakers is involved in determining the relation between input utterances and hence contributing to the verification process. We propose to create a set of artificial embeddings through the model training process. Utilizing the generated embeddings as auxiliary speakers, no extra data are required for the graph model in the verification stage. The proposed model is trained in an end-to-end manner within the whole system. Experiments are carried out with the Voxceleb datasets. The results indicate that involving auxiliary speakers with graph is effective to improve speaker verification performance.

Wei Liu, Tan Lee

Confidence measure is a performance index of particular importance for automatic speech recognition (ASR) systems deployed in real-world scenarios. In the present study, utterance-level neural confidence measure (NCM) in end-to-end automatic speech recognition (E2E ASR) is investigated. The E2E system adopts the joint CTC-attention Transformer architecture. The prediction of NCM is formulated as a task of binary classification, i.e., accept/reject the input utterance, based on a set of predictor features acquired during the ASR decoding process. The investigation is focused on evaluating and comparing the efficacies of predictor features that are derived from different internal and external modules of the E2E system. Experiments are carried out on children speech, for which state-of-the-art ASR systems show less than satisfactory performance and robust confidence measure is particularly useful. It is noted that predictor features related to acoustic information of speech play a more important role in estimating confidence measure than those related to linguistic information. N-best score features show significantly better performance than single-best ones. It has also been shown that the metrics of EER and AUC are not appropriate to evaluate the NCM of a mismatched ASR with significant performance gap.

Yuzhong Wu, Tan Lee

Acoustic scene classification (ASC) aims to identify the type of scene (environment) in which a given audio signal is recorded. The log-mel feature and convolutional neural network (CNN) have recently become the most popular time-frequency (TF) feature representation and classifier in ASC. An audio signal recorded in a scene may include various sounds overlapping in time and frequency. The previous study suggests that separately considering the long-duration sounds and short-duration sounds in CNN may improve ASC accuracy. This study addresses the problem of the generalization ability of acoustic scene classifiers. In practice, acoustic scene signals' characteristics may be affected by various factors, such as the choice of recording devices and the change of recording locations. When an established ASC system predicts scene classes on audios recorded in unseen scenarios, its accuracy may drop significantly. The long-duration sounds not only contain domain-independent acoustic scene information, but also contain channel information determined by the recording conditions, which is prone to over-fitting. For a more robust ASC system, We propose a robust feature learning (RFL) framework to train the CNN. The RFL framework down-weights CNN learning specifically on long-duration sounds. The proposed method is to train an auxiliary classifier with only long-duration sound information as input. The auxiliary classifier is trained with an auxiliary loss function that assigns less learning weight to poorly classified examples than the standard cross-entropy loss. The experimental results show that the proposed RFL framework can obtain a more robust acoustic scene classifier towards unseen devices and cities.

Guangyan Zhang, Ying Qin, Daxin Tan et al.

This paper describes a novel design of a neural network-based speech generation model for learning prosodic representation.The problem of representation learning is formulated according to the information bottleneck (IB) principle. A modified VQ-VAE quantized layer is incorporated in the speech generation model to control the IB capacity and adjust the balance between reconstruction power and disentangle capability of the learned representation. The proposed model is able to learn word-level prosodic representations from speech data. With an optimized IB capacity, the learned representations not only are adequate to reconstruct the original speech but also can be used to transfer the prosody onto different textual content. Extensive results of the objective and subjective evaluation are presented to demonstrate the effect of IB capacity control, the effectiveness, and potential usage of the learned prosodic representation in controllable neural speech generation.

Daxin Tan, Liqun Deng, Yu Ting Yeung et al.

This paper presents the design, implementation and evaluation of a speech editing system, named EditSpeech, which allows a user to perform deletion, insertion and replacement of words in a given speech utterance, without causing audible degradation in speech quality and naturalness. The EditSpeech system is developed upon a neural text-to-speech (NTTS) synthesis framework. Partial inference and bidirectional fusion are proposed to effectively incorporate the contextual information related to the edited region and achieve smooth transition at both left and right boundaries. Distortion introduced to the unmodified parts of the utterance is alleviated. The EditSpeech system is developed and evaluated on English and Chinese in multi-speaker scenarios. Objective and subjective evaluation demonstrate that EditSpeech outperforms a few baseline systems in terms of low spectral distortion and preferred speech quality. Audio samples are available online for demonstration https://daxintan-cuhk.github.io/EditSpeech/ .

Si-Ioi Ng, Cymie Wing-Yee Ng, Jingyu Li et al.

Speech sound disorder (SSD) refers to a type of developmental disorder in young children who encounter persistent difficulties in producing certain speech sounds at the expected age. Consonant errors are the major indicator of SSD in clinical assessment. Previous studies on automatic assessment of SSD revealed that detection of speech errors concerning short and transitory consonants is less satisfactory. This paper investigates a neural network based approach to detecting consonant errors in disordered speech using consonant-vowel (CV) diphone segment in comparison to using consonant monophone segment. The underlying assumption is that the vowel part of a CV segment carries important information of co-articulation from the consonant. Speech embeddings are extracted from CV segments by a recurrent neural network model. The similarity scores between the embeddings of the test segment and the reference segments are computed to determine if the test segment is the expected consonant or not. Experimental results show that using CV segments achieves improved performance on detecting speech errors concerning those "difficult" consonants reported in the previous studies.

Shuiyang Mao, P. C. Ching, Tan Lee

Despite the widespread utilization of deep neural networks (DNNs) for speech emotion recognition (SER), they are severely restricted due to the paucity of labeled data for training. Recently, segment-based approaches for SER have been evolving, which train backbone networks on shorter segments instead of whole utterances, and thus naturally augments training examples without additional resources. However, one core challenge remains for segment-based approaches: most emotional corpora do not provide ground-truth labels at the segment level. To supervisely train a segment-based emotion model on such datasets, the most common way assigns each segment the corresponding utterance's emotion label. However, this practice typically introduces noisy (incorrect) labels as emotional information is not uniformly distributed across the whole utterance. On the other hand, DNNs have been shown to easily over-fit a dataset when being trained with noisy labels. To this end, this work proposes a simple and effective deep self-learning (DSL) framework, which comprises a procedure to progressively correct segment-level labels in an iterative learning manner. The DSL method produces dynamically-generated and soft emotion labels, leading to significant performance improvements. Experiments on three well-known emotional corpora demonstrate noticeable gains using the proposed method.

Daxin Tan, Hingpang Huang, Guangyan Zhang et al.

This paper presents the CUHK-EE voice cloning system for ICASSP 2021 M2VoC challenge. The challenge provides two Mandarin speech corpora: the AIShell-3 corpus of 218 speakers with noise and reverberation and the MST corpus including high-quality speech of one male and one female speakers. 100 and 5 utterances of 3 target speakers in different voice and style are provided in track 1 and 2 respectively, and the participants are required to synthesize speech in target speaker's voice and style. We take part in the track 1 and carry out voice cloning based on 100 utterances of target speakers. An end-to-end voicing cloning system is developed to accomplish the task, which includes: 1. a text and speech front-end module with the help of forced alignment, 2. an acoustic model combining Tacotron2 and DurIAN to predict melspectrogram, 3. a Hifigan vocoder for waveform generation. Our system comprises three stages: multi-speaker training stage, target speaker adaption stage and target speaker synthesis stage. Our team is identified as T17. The subjective evaluation results provided by the challenge organizer demonstrate the effectiveness of our system. Audio samples are available at our demo page: https://daxintan-cuhk.github.io/CUHK-EE-system-M2VoC-challenge/ .

Xurong Xie, Xunying Liu, Tan Lee et al.

A key task for speech recognition systems is to reduce the mismatch between training and evaluation data that is often attributable to speaker differences. Speaker adaptation techniques play a vital role to reduce the mismatch. Model-based speaker adaptation approaches often require sufficient amounts of target speaker data to ensure robustness. When the amount of speaker level data is limited, speaker adaptation is prone to overfitting and poor generalization. To address the issue, this paper proposes a full Bayesian learning based DNN speaker adaptation framework to model speaker-dependent (SD) parameter uncertainty given limited speaker specific adaptation data. This framework is investigated in three forms of model based DNN adaptation techniques: Bayesian learning of hidden unit contributions (BLHUC), Bayesian parameterized activation functions (BPAct), and Bayesian hidden unit bias vectors (BHUB). In the three methods, deterministic SD parameters are replaced by latent variable posterior distributions for each speaker, whose parameters are efficiently estimated using a variational inference based approach. Experiments conducted on 300-hour speed perturbed Switchboard corpus trained LF-MMI TDNN/CNN-TDNN systems suggest the proposed Bayesian adaptation approaches consistently outperform the deterministic adaptation on the NIST Hub5'00 and RT03 evaluation sets. When using only the first five utterances from each speaker as adaptation data, significant word error rate reductions up to 1.4% absolute (7.2% relative) were obtained on the CallHome subset. The efficacy of the proposed Bayesian adaptation techniques is further demonstrated in a comparison against the state-of-the-art performance obtained on the same task using the most recent systems reported in the literature.

Man-Ling Sung, Tan Lee

Spoken term discovery from untranscribed speech audio could be achieved via a two-stage process. In the first stage, the unlabelled speech is decoded into a sequence of subword units that are learned and modelled in an unsupervised manner. In the second stage, partial sequence matching and clustering are performed on the decoded subword sequences, resulting in a set of discovered words or phrases. A limitation of this approach is that the results of subword decoding could be erroneous, and the errors would impact the subsequent steps. While Siamese/Triplet network is one approach to learn segment representations that can improve the discovery process, the challenge in spoken term discovery under a complete unsupervised scenario is that training examples are unavailable. In this paper, we propose to generate training examples from initial hypothesized sequence clusters. The Siamese/Triplet network is trained on the hypothesized examples to measure the similarity between two speech segments and hereby perform re-clustering of all hypothesized subword sequences to achieve spoken term discovery. Experimental results show that the proposed approach is effective in obtaining training examples for Siamese and Triplet networks, improving the efficacy of spoken term discovery as compared with the original two-stage method.

Si-Ioi Ng, Wei Liu, Zhiyuan Peng et al.

This technical report describes our submission to the 2021 SLT Children Speech Recognition Challenge (CSRC) Track 1. Our approach combines the use of a joint CTC-attention end-to-end (E2E) speech recognition framework, transfer learning, data augmentation and development of various language models. Procedures of data pre-processing, the background and the course of system development are described. The analysis of the experiment results, as well as the comparison between the E2E and DNN-HMM hybrid system are discussed in detail. Our system achieved a character error rate (CER) of 20.1% in our designated test set, and 23.6% in the official evaluation set, which is placed at 10-th overall.

Daxin Tan, Tan Lee

This paper presents a novel design of neural network system for fine-grained style modeling, transfer and prediction in expressive text-to-speech (TTS) synthesis. Fine-grained modeling is realized by extracting style embeddings from the mel-spectrograms of phone-level speech segments. Collaborative learning and adversarial learning strategies are applied in order to achieve effective disentanglement of content and style factors in speech and alleviate the "content leakage" problem in style modeling. The proposed system can be used for varying-content speech style transfer in the single-speaker scenario. The results of objective and subjective evaluation show that our system performs better than other fine-grained speech style transfer models, especially in the aspect of content preservation. By incorporating a style predictor, the proposed system can also be used for text-to-speech synthesis. Audio samples are provided for system demonstration https://daxintan-cuhk.github.io/pl-csd-speech .

Man-Ling Sung, Siyuan Feng, Tan Lee

The present study tackles the problem of automatically discovering spoken keywords from untranscribed audio archives without requiring word-by-word speech transcription by automatic speech recognition (ASR) technology. The problem is of practical significance in many applications of speech analytics, including those concerning low-resource languages, and large amount of multilingual and multi-genre data. We propose a two-stage approach, which comprises unsupervised acoustic modeling and decoding, followed by pattern mining in acoustic unit sequences. The whole process starts by deriving and modeling a set of subword-level speech units with untranscribed data. With the unsupervisedly trained acoustic models, a given audio archive is represented by a pseudo transcription, from which spoken keywords can be discovered by string mining algorithms. For unsupervised acoustic modeling, a deep neural network trained by multilingual speech corpora is used to generate speech segmentation and compute bottleneck features for segment clustering. Experimental results show that the proposed system is able to effectively extract topic-related words and phrases from the lecture recordings on MIT OpenCourseWare.

Shuiyang Mao, P. C. Ching, C. -C. Jay Kuo et al.

Categorical speech emotion recognition is typically performed as a sequence-to-label problem, i.e., to determine the discrete emotion label of the input utterance as a whole. One of the main challenges in practice is that most of the existing emotion corpora do not give ground truth labels for each segment; instead, we only have labels for whole utterances. To extract segment-level emotional information from such weakly labeled emotion corpora, we propose using multiple instance learning (MIL) to learn segment embeddings in a weakly supervised manner. Also, for a sufficiently long utterance, not all of the segments contain relevant emotional information. In this regard, three attention-based neural network models are then applied to the learned segment embeddings to attend the most salient part of a speech utterance. Experiments on the CASIA corpus and the IEMOCAP database show better or highly competitive results than other state-of-the-art approaches.

Shuiyang Mao, P. C. Ching, Tan Lee

Human emotional speech is, by its very nature, a variant signal. This results in dynamics intrinsic to automatic emotion classification based on speech. In this work, we explore a spectral decomposition method stemming from fluid-dynamics, known as Dynamic Mode Decomposition (DMD), to computationally represent and analyze the global utterance-level dynamics of emotional speech. Specifically, segment-level emotion-specific representations are first learned through an Emotion Distillation process. This forms a multi-dimensional signal of emotion flow for each utterance, called Emotion Profiles (EPs). The DMD algorithm is then applied to the resultant EPs to capture the eigenfrequencies, and hence the fundamental transition dynamics of the emotion flow. Evaluation experiments using the proposed approach, which we call EigenEmo, show promising results. Moreover, due to the positive combination of their complementary properties, concatenating the utterance representations generated by EigenEmo with simple EPs averaging yields noticeable gains.

Shuiyang Mao, P. C. Ching, Tan Lee

Human emotions are inherently ambiguous and impure. When designing systems to anticipate human emotions based on speech, the lack of emotional purity must be considered. However, most of the current methods for speech emotion classification rest on the consensus, e.g., one single hard label for an utterance. This labeling principle imposes challenges for system performance considering emotional impurity. In this paper, we recommend the use of emotional profiles (EPs), which provides a time series of segment-level soft labels to capture the subtle blends of emotional cues present across a specific speech utterance. We further propose the emotion profile refinery (EPR), an iterative procedure to update EPs. The EPR method produces soft, dynamically-generated, multiple probabilistic class labels during successive stages of refinement, which results in significant improvements in the model accuracy. Experiments on three well-known emotion corpora show noticeable gain using the proposed method.

Si-Ioi Ng, Tan Lee

Speech sound disorder (SSD) refers to the developmental disorder in which children encounter persistent difficulties in correctly pronouncing words. Assessment of SSD has been relying largely on trained speech and language pathologists (SLPs). With the increasing demand for and long-lasting shortage of SLPs, automated assessment of speech disorder becomes a highly desirable approach to assisting clinical work. This paper describes a study on automatic detection of phonological errors in Cantonese speech of kindergarten children, based on a newly collected large speech corpus. The proposed approach to speech error detection involves the use of a Siamese recurrent autoencoder, which is trained to learn the similarity and discrepancy between phone segments in the embedding space. Training of the model requires only speech data from typically developing (TD) children. To distinguish disordered speech from typical one, cosine distance between the embeddings of the test segment and the reference segment is computed. Different model architectures and training strategies are experimented. Results on detecting the 6 most common consonant errors demonstrate satisfactory performance of the proposed model, with the average precision value from 0.82 to 0.93.

Si-Ioi Ng, Cymie Wing-Yee Ng, Jiarui Wang et al.

This paper describes the design and development of CUCHILD, a large-scale Cantonese corpus of child speech. The corpus contains spoken words collected from 1,986 child speakers aged from 3 to 6 years old. The speech materials include 130 words of 1 to 4 syllables in length. The speakers cover both typically developing (TD) children and children with speech disorder. The intended use of the corpus is to support scientific and clinical research, as well as technology development related to child speech assessment. The design of the corpus, including selection of words, participants recruitment, data acquisition process, and data pre-processing are described in detail. The results of acoustical analysis are presented to illustrate the properties of child speech. Potential applications of the corpus in automatic speech recognition, phonological error detection and speaker diarization are also discussed.

Zhiyuan Peng, Siyuan Feng, Tan Lee

Speech signal is constituted and contributed by various informative factors, such as linguistic content and speaker characteristic. There have been notable recent studies attempting to factorize speech signal into these individual factors without requiring any annotation. These studies typically assume continuous representation for linguistic content, which is not in accordance with general linguistic knowledge and may make the extraction of speaker information less successful. This paper proposes the mixture factorized auto-encoder (mFAE) for unsupervised deep factorization. The encoder part of mFAE comprises a frame tokenizer and an utterance embedder. The frame tokenizer models linguistic content of input speech with a discrete categorical distribution. It performs frame clustering by assigning each frame a soft mixture label. The utterance embedder generates an utterance-level vector representation. A frame decoder serves to reconstruct speech features from the encoders'outputs. The mFAE is evaluated on speaker verification (SV) task and unsupervised subword modeling (USM) task. The SV experiments on VoxCeleb 1 show that the utterance embedder is capable of extracting speaker-discriminative embeddings with performance comparable to a x-vector baseline. The USM experiments on ZeroSpeech 2017 dataset verify that the frame tokenizer is able to capture linguistic content and the utterance embedder can acquire speaker-related information.

Siyuan Feng, Tan Lee

This research addresses the problem of acoustic modeling of low-resource languages for which transcribed training data is absent. The goal is to learn robust frame-level feature representations that can be used to identify and distinguish subword-level speech units. The proposed feature representations comprise various types of multilingual bottleneck features (BNFs) that are obtained via multi-task learning of deep neural networks (MTL-DNN). One of the key problems is how to acquire high-quality frame labels for untranscribed training data to facilitate supervised DNN training. It is shown that learning of robust BNF representations can be achieved by effectively leveraging transcribed speech data and well-trained automatic speech recognition (ASR) systems from one or more out-of-domain (resource-rich) languages. Out-of-domain ASR systems can be applied to perform speaker adaptation with untranscribed training data of the target language, and to decode the training speech into frame-level labels for DNN training. It is also found that better frame labels can be generated by considering temporal dependency in speech when performing frame clustering. The proposed methods of feature learning are evaluated on the standard task of unsupervised subword modeling in Track 1 of the ZeroSpeech 2017 Challenge. The best performance achieved by our system is $9.7\%$ in terms of across-speaker triphone minimal-pair ABX error rate, which is comparable to the best systems reported recently. Lastly, our investigation reveals that the closeness between target languages and out-of-domain languages and the amount of available training data for individual target languages could have significant impact on the goodness of learned features.

Siyuan Feng, Tan Lee

This study tackles unsupervised subword modeling in the zero-resource scenario, learning frame-level speech representation that is phonetically discriminative and speaker-invariant, using only untranscribed speech for target languages. Frame label acquisition is an essential step in solving this problem. High quality frame labels should be in good consistency with golden transcriptions and robust to speaker variation. We propose to improve frame label acquisition in our previously adopted deep neural network-bottleneck feature (DNN-BNF) architecture by applying the factorized hierarchical variational autoencoder (FHVAE). FHVAEs learn to disentangle linguistic content and speaker identity information encoded in speech. By discarding or unifying speaker information, speaker-invariant features are learned and fed as inputs to DPGMM frame clustering and DNN-BNF training. Experiments conducted on ZeroSpeech 2017 show that our proposed approaches achieve $2.4\%$ and $0.6\%$ absolute ABX error rate reductions in across- and within-speaker conditions, comparing to the baseline DNN-BNF system without applying FHVAEs. Our proposed approaches significantly outperform vocal tract length normalization in improving frame labeling and subword modeling.

Siyuan Feng, Tan Lee, Zhiyuan Peng

This study addresses the problem of unsupervised subword unit discovery from untranscribed speech. It forms the basis of the ultimate goal of ZeroSpeech 2019, building text-to-speech systems without text labels. In this work, unit discovery is formulated as a pipeline of phonetically discriminative feature learning and unit inference. One major difficulty in robust unsupervised feature learning is dealing with speaker variation. Here the robustness towards speaker variation is achieved by applying adversarial training and FHVAE based disentangled speech representation learning. A comparison of the two approaches as well as their combination is studied in a DNN-bottleneck feature (DNN-BNF) architecture. Experiments are conducted on ZeroSpeech 2019 and 2017. Experimental results on ZeroSpeech 2017 show that both approaches are effective while the latter is more prominent, and that their combination brings further marginal improvement in across-speaker condition. Results on ZeroSpeech 2019 show that in the ABX discriminability task, our approaches significantly outperform the official baseline, and are competitive to or even outperform the official topline. The proposed unit sequence smoothing algorithm improves synthesis quality, at a cost of slight decrease in ABX discriminability.

Yuzhong Wu, Tan Lee

Acoustic scene classification is the task of identifying the scene from which the audio signal is recorded. Convolutional neural network (CNN) models are widely adopted with proven successes in acoustic scene classification. However, there is little insight on how an audio scene is perceived in CNN, as what have been demonstrated in image recognition research. In the present study, the Class Activation Mapping (CAM) is utilized to analyze how the log-magnitude Mel-scale filter-bank (log-Mel) features of different acoustic scenes are learned in a CNN classifier. It is noted that distinct high-energy time-frequency components of audio signals generally do not correspond to strong activation on CAM, while the background sound texture are well learned in CNN. In order to make the sound texture more salient, we propose to apply the Difference of Gaussian (DoG) and Sobel operator to process the log-Mel features and enhance edge information of the time-frequency image. Experimental results on the DCASE 2017 ASC challenge show that using edge enhanced log-Mel images as input feature of CNN significantly improves the performance of audio scene classification.

Yuzhong Wu, Tan Lee

Audio classification is the task of identifying the sound categories that are associated with a given audio signal. This paper presents an investigation on large-scale audio classification based on the recently released AudioSet database. AudioSet comprises 2 millions of audio samples from YouTube, which are human-annotated with 527 sound category labels. Audio classification experiments with the balanced training set and the evaluation set of AudioSet are carried out by applying different types of neural network models. The classification performance and the model complexity of these models are compared and analyzed. While the CNN models show better performance than MLP and RNN, its model complexity is relatively high and undesirable for practical use. We propose two different strategies that aim at constructing low-dimensional embedding feature extractors and hence reducing the number of model parameters. It is shown that the simplified CNN model has only 1/22 model parameters of the original model, with only a slight degradation of performance.