Man-Machine Speech Communication: 14th National Conference, NCMMSC 2017, Lianyungang, China, October 11–13, 2017, Revised Selected Papers (Communications in Computer and Information Science Book 807)

This book constitutes the refereed proceedings of the 14th National Conference on Man-Machine Speech Communication, NCMMSC 2017, held in Lianyungang, China, in October 2017. The 13 revised full papers presented were carefully reviewed and selected from 39 submissions. The papers address issues such as challenging issues in speech recognition and enhancement, speaker and language recognition, speech synthesis, corpus and phonetic in speech technology, speech generation, speech analyzing and modelling, speech processing of ethnic minorities, speech emotion recognition and audio signal processing. Preface 6 Organization 7 Contents 10 Uyghur Word Stemming Based on Stem and Affix Features 12 Abstract 12 1 Introduction 12 2 Survey of Related Work 14 3 Implementation of an Uyghur Stemmer 15 3.1 A Stemming Framework for Uyghur Words 15 3.2 A Proposed Approach for an Uyghur Stemmer 16 3.3 Establishment of an Uyghur Stem Segmentation Corpus 17 3.3.1 Stem List 17 3.3.2 Suffix List 18 3.3.3 Training Corpus 19 3.4 Stemming a Given Uyghur Word 20 4 Experiments and Results 20 4.1 Experimental Settings 20 4.2 Analysis of Experiments and Results 21 4.2.1 Accuracy Based on the Standard Data 21 4.2.2 Limiting the Number of Singular Suffixes 21 4.2.3 Providing Uneven Weights to Stem and Suffix Limiting the Number of Singular Suffix 22 5 Conclusions 22 Acknowledgement 22 References 23 GLEU-Guided Multi-resolution Network for Short Text Conversation 24 1 Introduction 24 2 Model Architecture 26 2.1 Network Structure for Keywords-Sequence Generation 27 2.2 GLEU-Guided Policy Gradient Training 28 2.3 Response Generation 29 3 Experiments 29 3.1 Data Set 29 3.2 Training Details 30 3.3 Evaluation Methods 30 3.4 Results 30 3.5 Case Study 32 4 Conclusion 33 References 33 Applying Functional Partition in the Investigation of Lexical Tonal-Pattern Categories in an Under-Resourced Chinese Dialect 35 Abstract 35 1 Introduction 36 1.1 Two-Stage Semi-automatic Partition 36 2 Experiment 38 2.1 Corpus Preparation 38 2.2 Word-Wise Partitioning and Verification 39 2.3 Partitioning for Basic Tonal Patterns 40 3 Results 42 3.1 Word-Wise Partitioning 42 3.2 Optimized and Adjusted General Partitioning Results 43 3.3 Discussion and Conclusion 45 Acknowledgements 45 References 45 Typology of Convergences and Divergences of English Monophthongs by EFL Learners from Guanhua Regions 47 Abstract 47 1 Introduction 47 2 Methodology 48 2.1 The Vowel Inventory of Guanhua Dialect 48 2.2 Materials 49 2.3 Subjects 49 2.4 Recording 49 2.5 Data Analysis 49 3 Findings 50 3.1 Spectral Properties 50 3.2 Transfer from Dialects 51 3.3 Transfer from Dialects 53 4 Discussions 55 5 Conclusions 55 Acknowledgements 56 References 56 Perception of English Phonemes by Chinese College Students 58 Abstract 58 1 Introduction 58 2 Method 60 2.1 Participants 60 2.2 Stimuli 60 2.3 Procedure 61 3 Results 61 3.1 Vowels and Consonants 61 3.2 Vowels 62 3.3 Consonants 64 4 Discussion 65 Acknowledgments 67 References 67 Collaborative Learning for Language and Speaker Recognition 69 1 Introduction 69 2 Related Work 70 3 Multi-task RNN and Collaborative Learning 71 3.1 Basic Single-Task Model 71 3.2 Multi-task Recurrent Model 72 3.3 Model Training 74 4 Experiments 74 4.1 Data 74 4.2 LRE and SRE Baselines 75 4.3 Collaborative Learning 77 5 Conclusions 78 References 78 HelloNPU: A Corpus for Small-Footprint Wake-Up Word Detection Research 81 Abstract 81 1 Introduction 81 2 Related Work 82 3 The Corpus 83 4 Deep KWS System 84 4.1 Feature Extraction 84 4.2 Deep Neural Network 85 4.3 Posterior Handling 85 5 Data Augmentation 86 6 Experimental Results 87 7 Conclusion 89 References 89 Multi-task Learning in Prediction and Correction for Low Resource Speech Recognition 91 Abstract 91 1 Introduction 91 2 PAC-MTL-CLDNN Combined Architecture 93 2.1 Model Structure 93 2.2 Multilingual CNN Model 93 2.3 Multitask Learning DNN Model 94 2.4 Multi-scale Features 94 2.5 Joint Acoustic Modeling with UPS and UGS 94 3 Experiment 95 3.1 Database and Task 95 3.2 Setup 95 3.3 Results and Evaluation 96 4 Conclusion 97 Acknowledgements 98 References 98 Acoustic Model Compression with Knowledge Transfer 100 Abstract 100 1 Introduction 100 2 Related Work 101 3 Proposed Compression Method 102 4 Framework of the Proposed Method 103 5 Experiments 104 5.1 Mandarin Corpus: RASC863 104 5.2 English Corpus: AMI 106 6 Discussions 107 7 Conclusions 107 Acknowledgements 108 References 108 Mongolian Text-to-Speech System Based on Deep Neural Network 110 1 Introduction 110 2 Mongolian TTS System Based on DNN 111 2.1 Training Part 111 2.2 Synthesis Part 113 3 Experiments and Results 115 3.1 Dataset 115 3.2 Experiments Setup 115 3.3 Evaluation 116 4 Conclusions 118 References 118 Using Mandarin Training Corpus to Realize a Mandarin-Tibetan Cross-Lingual Emotional Speech Synthesis 120 1 Introduction 120 2 Framework of Mandarin-Tibetan Cross-Lingual Emotional Speech Synthesis 121 3 Context-Dependent Labels 123 4 Experiments 124 4.1 Experimental Conditions 124 4.2 Language Similarity 125 4.3 Speech Quality 126 4.4 Emotion Similarity 127 4.5 Objective Evaluation 129 5 Conclusions 130 References 131 Emotion Recognition Using Support Vector Machine and Deep Neural Network 133 1 Introduction 133 2 Methods 134 2.1 LLD-DNN Subsystem 135 2.2 LLD-SVM Subsystem 135 2.3 DNN-SVM Subsystem 136 2.4 Combining Subsystems 137 3 Experiments 138 3.1 Experiment Settings 138 3.2 Results 139 4 Conclusion 141 References 141 Distance-Dependent Modeling of Head-Related Transfer Functions Based on Spherical Fourier-Bessel Transform 143 1 Introduction 143 2 Spherical Harmonic-Based Model 145 3 Proposed DSHM 145 3.1 HRTFs Preprocessing 146 3.2 Spherical Fourier-Bessel-Based Transform 146 3.3 Least Square Modeling 147 3.4 Continuous Construction of HRTFs 148 4 Performance Evaluation 149 5 Conclusions 151 References 151 Author Index 153 Front Matter ....Pages I-XII Uyghur Word Stemming Based on Stem and Affix Features (Hankiz Yilahun, Sediyegvl Enwer, Askar Hamdulla)....Pages 1-12 GLEU-Guided Multi-resolution Network for Short Text Conversation (Xuan Liu, Kai Yu)....Pages 13-23 Applying Functional Partition in the Investigation of Lexical Tonal-Pattern Categories in an Under-Resourced Chinese Dialect (Junru Wu, Yiya Chen, Vincent J. van Heuven, Niels O. Schiller)....Pages 24-35 Typology of Convergences and Divergences of English Monophthongs by EFL Learners from Guanhua Regions (Yuan Jia, Yu Wang, Aijun Li, Dawei Song, Liang Xu)....Pages 36-46 Perception of English Phonemes by Chinese College Students (Yanqin Feng, Hao Yan, Liangkai Zhai)....Pages 47-57 Collaborative Learning for Language and Speaker Recognition (Lantian Li, Zhiyuan Tang, Dong Wang, Andrew Abel, Yang Feng, Shiyue Zhang)....Pages 58-69 HelloNPU: A Corpus for Small-Footprint Wake-Up Word Detection Research (Senmao Wang, Jingyong Hou, Lei Xie, Yufeng Hao)....Pages 70-79 Multi-task Learning in Prediction and Correction for Low Resource Speech Recognition (Danish Bukhari, Jiangyan Yi, Zhengqi Wen, Bin Liu, Jianhua Tao)....Pages 80-88 Acoustic Model Compression with Knowledge Transfer (Jiangyan Yi, Jianhua Tao, Zhengqi Wen, Ya Li, Hao Ni)....Pages 89-98 Mongolian Text-to-Speech System Based on Deep Neural Network (Rui Liu, Feilong Bao, Guanglai Gao, Yonghe Wang)....Pages 99-108 Using Mandarin Training Corpus to Realize a Mandarin-Tibetan Cross-Lingual Emotional Speech Synthesis (Peiwen Wu, Hongwu Yang, Zhenye Gan)....Pages 109-121 Emotion Recognition Using Support Vector Machine and Deep Neural Network (Ruinian Chen, Ying Zhou, Yanmin Qian)....Pages 122-131 Distance-Dependent Modeling of Head-Related Transfer Functions Based on Spherical Fourier-Bessel Transform (Xiaoke Qi, Jianhua Tao)....Pages 132-141 Back Matter ....Pages 143-143