DECLARATION OF AUTHORSHIP I, Pham Van Dong, declare that the dissertation titled “Speech Synthesis for Low- Resourced Languages based on Adaptation Approach: Application to Muong Language” has been entirely composed by myself. I assure you of some points as follows: This work was done wholly or mainly while in candidature for a Ph. research degree at Hanoi University of Science and Technology. The work has not been submitted for any other degree or qualifications at Hanoi University of Science and Technology or any other institution.
Appropriate acknowledgment has been given within this dissertation, where reference has been made to the published work of others. The dissertation submitted is my own, except where work in the collaboration has been included. The collaborative contributions have been indicated. Hanoi, December 8, 2023 Ph.
Student Pham Van Dong ADVISORS 1. Mac Dang Khoa 2. Tran Do Dat i ACKNOWLEDGMENT Foremost, I would like to express my most sincere and deepest gratitude to my thesis advisors Dr. Mạc Đăng Khoa (Speech Communication Department, MultiLab at MICA), Prof.
TRẦN Đỗ Đạt (The Ministry of Science and Technology, Vietnam) for their continuous support and guidance during my Ph. program, and for providing me with such a severe and inspiring research environment. I am grateful to Dr. Mạc Đăng Khoa for his excellent mentorship, caring, patience, and immense Text-To-Speech (TTS) knowledge.
His advice helped me in all the research and writing of this thesis. I am very thankful to Prof. Đạt for shaping my thesis at the beginning and for their enthusiasm and encouragement. Trần Đỗ Đạt substantially facilitated my Ph.
research, especially when I was a freshman on speech processing and TTS, with his valuable comments on Vietnamese and Muong TTS. I thank all MICA members for their help during my Ph. My sincere thanks to Dr. Nguyen Viet Son, Assoc.
Dao Trung Kien and Dr. Do Thi Ngoc Diep for giving me much support and valuable advice. Thanks to Nguyen Van Thinh, Nguyen Tien Thanh, Dang Thanh Mai, and Vu Thi Hai Ha for their help. I want to thank my Hanoi University of Mining and Geology colleagues for all their support during my Ph.
Special thanks to my family for understanding my hours glued to the computer screen. Hanoi, December 8, 2023 Ph. Student ii ABSTRACT Text-to-speech (TTS) synthesis is the automatic conversion of text into speech. Typically, building high-quality voiceovers requires collecting tens of hours of the voice of a professional speaker with a high-quality microphone.
There are about 7,000 languages spoken worldwide, but only a few languages, such as English, Spanish, Mandarin, and Japanese, are used in good TTS. With so-called "low-resourced languages" or even languages that are not yet written, these languages do not have TTS. Thus, to apply TTS technology to low-resourced language, it is necessary to study other TTS methods. In Vietnam, Vietnamese is the mother tongue and is used the most.
The Muong is a group of the language spoken by the Muong people of Vietnam. They are in the Austroasiatic language family and are closely related to Vietnamese, and Muong is also one of the five ethnic groups with the largest population. However, Muong still needs an official script, a typical representative of the low-resourced language in Vietnam. Therefore, researching TTS technologies to create TTS for the Muong language is challenging.
In the first part of this thesis, we do an overview of TTS. Researching the phonetics of Vietnamese and Muong languages, the thesis has also researched and published some tools to support TTS technology for Vietnamese and Muong languages. In the rest of the thesis, we conduct various experiments in creating TTS for low-resourced language; specifically, we experiment with the Muong language. We focus on two main low-resourced language groups: Written: We use emulating to simulate the reading of the Muong language using Vietnamese TTS and cross-lingual adaptation transfer-learning.
Unwritten: We experiment with adaptation in two directions. The first is to create Muong speech synthesis directly from Vietnamese Text and Muong voice. The second is to create Muong speech synthesis from translation through intermediate representation We hope our findings can serve as an impetus to develop speech synthesis for low-resourced languages worldwide and contribute to the basis for speech synthesis development for 53 ethnic minority languages in Viet Nam. Hanoi, December 8, 2023 Ph.
Student iii CONTENT DECLARATION OF AUTHORSHIP. VIII LIST OF TABLES. X LIST OF FIGURES. 1 PART 1 : BACKGROUND AND RELATED WORKS.
OVERVIEW OF SPEECH SYNTHESIS AND SPEECH SYNTHESIS FOR LOW-RESOURCED LANGUAGE. Overview of speech synthesis. Evolution of TTS methods over time. TTS using unit-selection method.
Statistical parameter speech synthesis. Speech synthesis using deep neural networks. Neural speech synthesis. Speech synthesis for low-resourced languages.
TTS using emulating input approach. TTS using the polyglot approach. Speech synthesis for low-resourced language using the adaptation approach. Neural translation model.
Attention in neural machine translation. Statistical machine translation based on phrase. Statistical machine translation problem based on phrase. Translation model and language model.
Decode the input sentence in the translation system. Model for building a statistical translation system. Machine translation through intermediate representation. Speech translation for unwritten low-resourced languages.
Speech synthesis evaluation metrics. Mean Opinion Score (MOS). Mel Cepstral Distortion (MCD). MCD with Dynamic Time Warping (MCD – DTW).
Analysis of variance (Anova). VIETNAMESE AND MUONG LANGUAGE. History of Vietnamese. Vietnamese phonetic system.
Vietnamese syllabus structure. Vietnamese phonetic system. Vietnamese tone system. Overview of Muong people and Muong language.
Viet Muong group. Muong written script. Muong phonetics system. Muong syllable structure.
Muong phoneme system. Muong tone system. Comparison between Vietnamese and Muong. Dicussion and proposal approach.
60 PART 2 : SPEECH SYNTHESIS FOR MUONG AS A WRITTEN LANGUAGE. EMULATING OF THE MUONG TTS BASED ON INPUT TRANSFORMATION OF THE VIETNAMESE TTS. Muong emulating IPA module. Analysis by ANOVA method.
MOS analysis by ANOVA. Intelligibility analysis by ANOVA. CROSS-LINGUAL TRANSFER LEARNING FOR MUONG SPEECH SYNTHESIS. Muong Project‘s data.
Muong fine-tuning data. Graphemes to phonemes. Training the pretrained model using Vietnamese dataset. Finetuned TTS model on Muong datasets.
MOS analysis by ANOVA. 94 PART 3 : SPEECH SYNTHESIS FOR MUONG AS AN UNWRITTEN LANGUAGE. GENERATE UNWRITTEN LOW-RESOURCED LANGUAGE’S SPEECH DIRECTLY FROM RICH-RESOURCE LANGUAGE’S TEXT. Training the speech synthesis system.
MOS analysis by ANOVA. ANOVA analysis in Muong Bi speech synthesis. ANOVA analysis in Muong Tan Son speech synthesis. SPEECH SYNTHESIS FOR UNWRITTEN LOW-RESOURCED LANGUAGE USING INTERMEDIATE REPRESENTATION.
Text to phone translation. Phone to Sound Conversion. Evaluation in Muong Bi and Muong Tan Son. MOS analysis by ANOVA.
ANOVA analysis in Muong Bi speech synthesis. ANOVA analysis in Muong Tan Son speech synthesis. Conclusion and comparison. 128 CONCLUSION AND FUTURE WORKS.
Vietnamese and Muong phonetic. Muong Vietnamese phone mapping. Information of Muong volunteers who participated in the assessment. Speech signal samples of the Muong TTS in chapter 5.
12 vii ABBREVIATIONS Expansion Explanation Abbreviation CART Classification And Regression Tree F0 Fundamental Frequency HMM Hidden Markov Model HTK Hidden markov model A portable toolkit for building and ToolKit manipulating hidden Markov models HTS HMM-based speech synthesis IPA International Phonetic Alphabet MARY Modular Architecture for (TTS) Research on speech sYnthesis MFCC Mel Frequency Cepstral Coefficents ML Maximum Likelihood MLSA Mel Log Spectrum Approximation MOS Mean Opinion Score MSD- Multi-Space probability HMM Distribution HMM NLP Natural Language Processing OCR Optical Character Recognition POS Part-Of-Speech Word class or a lexical category PP Prepositional Phrase PSOLA Pitch Synchronous OverLap and Add SAMPA Speech Assessment Methods Phonetic Alphabet SPTK Speech signal Processing ToolKit SSML Speech Synthesis Markup Language TD- Time-Domain Pitch PSOLA Synchronous OverLap and Add TTS Text-To-Speech VNSP VNSpeechCorpus for synthesis WEKA Waikato Environment for A collection of machine learning Knowledge algorithms for data mining tasks: Analysis X-SAMPA Extended Speech Assessment Methods Phonetic Alphabet XML eXtensible Markup Language PLP Perceptual Linear Prediction viii G2P Grapheme to Phoneme ANOVA Analysis of Variance DNN Deep Neural Network ANN Artificial Neural Network LPC Linear Predictive Coding EM Expectation Maximization Algorithm MLE Maximum Likelihood Network PTN Phonetic Transformation Network CNN Convolutional Neural Network NMT Neural Machine Translation SMT Statistical Machine Translation RNN Recurrent Neural Network GRU Gated Recurrent Unit DTW Dynamic Time Warping MCD Mel Ceptral Distortion Argmax Arguments of the maxima Argmax is an operation that finds the argument that gives the maximum value from a target function Log Logarith ̅ Sample mean p(e | f) Conditional Probability Pi Product of a sequence of numbers Sigma Factor H0 Null Hypothesis ix LIST OF TABLES Table 2.1 Vietnamese syllabus structure [94] .2 Vietnamese syllabus structure [96] .3 Vietnamese syllables based on structure .4 Hanoi Vietnamese inital consonants .5 The letter of initial consonant .6 Hanoi Vietnamese final consonant .7 Tone of Hanoi Vietnamese [108] .8 Muong syllabic structure .9 Muong final sound system .10 Muong Hoa Binh tone system [115] .11 Muong Bi and Muong Tan Son Tone .12 Muong and Vietnamese phonetic comparison (orthography in normal, IPA in italic; Vi: Vietnamese; Mb: Muong Bi ; Mts : Muong Tan Son) .13 Comparing the tone of Vietnamese with Muong Tan Son and Muong Bi .1 Muong G2P Result Sample .2 Examples of applying transformation rules to convert the Muong text into input text for Vietnamese TTS. Testing material for emulating tone. Testing material for emulating phone (the concerning phonemes in bold). Testing material for remaining phonemes .6 ANOVA Results for MOS Test.7 ANOVA Results for Intelligibility Test .1 Parameters of acoustic model .2 Vietnamese dataset information .3 Muong recorded data .4 The Muong split data set .5 Parameter for optimizer .6 Value of parameters when training Hifigan model .7 The specifications of the in-domain and out-domain test sets .8 Test set samples .10 ANOVA Results for in-domain MOS Test .11 ANOVA Results for out-domain MOS Test .12 ANOVA Results for in/out domain MOS Test .2 TTS evaluation with in-domain test set .3 TTS evaluation with out-domain test set .4 ANOVA Results for in-domain MOS Test for Muong Bi .5 ANOVA Results for out-domain MOS Test for Muong Bi .6 ANOVA Results for Muong Bi in/out domain MOS Test .7 ANOVA Results for in-domain MOS Test for Muong Tan Son .8 ANOVA Results for out-domain MOS Test for Muong Tan Son .9 ANOVA Results for Muong Tan Son in/out domain MOS Test.1 Examples of labeling Vietnamese text into an intermediate representation of Muong Bi and Muong Tan Son phonemes.2 Text information of Muong language datasets .3 TTS evaluation with in-domain test set .4 TTS evaluation with out-domain test set .5 ANOVA Results for in-domain MOS Test for Muong Bi .6 ANOVA Results for out-domain MOS Test for Muong Bi .7 ANOVA Results for Muong Bi in/out domain MOS Test .8 ANOVA Results for in-domain MOS Test for Muong Tan Son .9 ANOVA Results for out-domain MOS Test for Muong Tan Son .10 ANOVA Results for Muong Tan Son in/out domain MOS Test.2 The Muong initial consonant .3 Muong vowels system .4 The correspondences between Vietnamese and Muong in 12 words refer to the human body parts [137] .8 Muong Vietnamese phone mapping.9 Muong Hoa Binh volunteers .10 Muong Phu Tho volunteers .10 xi LIST OF FIGURES Figure 1.
Basic system architecture of a TTS system [22] .2 Neural TTS architecture [3]. General and clustering-based unit-selection scheme: Solid lines represent target costs and dashed lines represent concatenation costs [13]. Core architecture of HMM-based speech synthesis system [25]. General HMM-based synthesis scheme [13, p.
A speech synthesis framework based on a DNN [29] .7 Encoder and Decoder diagram in Seq2Seq model .8 Char2Wav model [23].9 Model of the Tacotron synthesis system [24] .10 Block diagram of the Tacotron 2 system architecture [25] .11 Scheme of a HMM-based polyglot synthesizer [48] .