MINISTRY OF EDUCATION AND TRAINING HANOI UNIVERSITY OF SCIENCE AND TECHNOLOGY ONOAId ONVIT NAANDN Nguyen Ilang Phuong MODELING DYNAMIC ACOUSTIC FEATURE OF SPEECH FOR VIETNAMESE SPEECH RECOGNITION AND APPLICATION FOR ANALYZING VOWEL — TO —- VOWEL TRANSITIONS SONATOS WALNdWO3 MO HÌNH DẶC TÍNH ÂM HỌC DONG CLIO NE DANG TUNG. NOL TIENG VIET VA UNG DỰNG CHO VIỆC PHAN TicH SY CHUYEN LIÊP NGUYEN AM — NGUYEN AM MASTER THESIS OF SCIENCE COMPUTER SCIENCE Y9I0E Hanoi — 2018 MINISTRY OF EDUCATION AND TRAINING HANOI UNIVERSITY OF SCIENCE AND TECHNOLOGY Nguyen Iang Phuong MODELING DYNAMIC ACOUSTIC FEATURE OF SPEECH FOR VIETNAMESE SPEECH RECOGNITION AND APPLICATION FOR, ANALYZING VOWEL — TO-— VOWEL TRANSITIONS. MO HINH DAC TINH AM HOC BONG CHO NHAN DANG TIENG NOI TIENG VIỆT VA UNG DUNG CHO VIEC PHAN TÍCH SỰ CHUYỀN TIẾP NGUYÊN ÂM ~ NGUYÊN ÂM Specialty: Computer Science International Rescarch Institute MICA MASTER THESIS OF SCIENCE COMPUTER SCIENCE SUPERVISOR: Prof. Eric Castelli Dr.
Nguyen Viet Son Hanoi — 2018 DECLARATION OF AUTIIORSINP 1, NGUYEN Hang Phuong, declare that this thesis tiled, “Modeling dynamic acoustic feature of speech for Viemamese speech recognition and application for analyzing Vowel to — Vowel transitions” and the work presented in il are my own. T confirm that ‘This work was done wholly or mainly while in candidature for a research degree at this University ‘Where any part of this thesis has previonsly been submitted for a degree or any other qualification at this University or any other institution, this has been olearly stated, Whore T have consulted the publishad work of others, this is always clearly athibuted. Whore [have quoted from the work of others, the source is always given. With the exception of such quotations, this thesis is entirely my own work.
Thave acknowledged all main sources of help. Whore the thesis is based on work donc by myself jointly with others, I have made clear exactly what was done by others and what | have contributed myself, Signed: Date: 3.2 /ao/ and /auw’ transition.3 đai and /ea/ transitiơn.4 /oa/ and /ua/ transifiơn.2 New proposal approach to automatic SSC angles computation.1 Definition and Antomatic SSCF angles calculation.2 Analysis SSCF angles in Vowel- Vowel transition 5? 3.1 /aU/ and /ae/ transiHIOH.2 /ao/ and /aư/ transifion.3 fia! and /ca/ transiliOn.4 foa/ and Aua/ transition - - 66 3.3 Conckusion of chapter 3. CHAPTER 4: APPLY SSCF ANGLES TO A SPEECH RECOGNITION TOOLKIT, KALDI 74 4.1 An overview about Katdl— an open source for speech recaguition.2 Balanced speaker experiments on Kaldt.1 Using MECC featues.2 Using SSCF anglss.3 Unbalanced speaker experiments on Kaldi using SSCF angles.4 Conclusion of chapter 4. CIIAPTER S$: CONCLUSION AND FUTURE WORK, - - 82 5.
ssessessesessssrsesessssniesseensansessensaniersaneaecsanionsse 85 APPENDIX: - - 8&7 ACKNOWLEDGEMENTS It is an honor for me to be here to write thankful words to those who have been supporting, guiding and inspiriting me from the moment, when I started my work in International Research Institute MICA until now. I owe my deepest gratitude to my supervisors, Prof, Eric Castelli and Dr. Nguyen Viet Son. Their expertise, undorstaniting and gencrous guidance mads il possible to work in a now topic for me, They have made available their support in a number of ways to find out the solution to my works.
It is a pleasure to work with them, Special thanks to Dr. Mac Dang Khoa, Dr. Do ‘Thi Ngoc Diep, Dr. Nguyen Cong Phuong and all of members in the Speech Communication Department for their guidance which help me a lot in how to study and to do research in right way, and also the valuable advices for my works.
Finally, this thesis would not have been possible if there were no encouragement from my family and friends. Their words give me power in order to overcome all the embarrassment, discouragament and other difficultics. Thanks for everything helping ms to got this day. Hanoi, 23/03/2018 Nguyen Hang Phuong ACKNOWLEDGEMENTS It is an honor for me to be here to write thankful words to those who have been supporting, guiding and inspiriting me from the moment, when I started my work in International Research Institute MICA until now.
I owe my deepest gratitude to my supervisors, Prof, Eric Castelli and Dr. Nguyen Viet Son. Their expertise, undorstaniting and gencrous guidance mads il possible to work in a now topic for me, They have made available their support in a number of ways to find out the solution to my works. It is a pleasure to work with them, Special thanks to Dr.
Mac Dang Khoa, Dr. Do ‘Thi Ngoc Diep, Dr. Nguyen Cong Phuong and all of members in the Speech Communication Department for their guidance which help me a lot in how to study and to do research in right way, and also the valuable advices for my works. Finally, this thesis would not have been possible if there were no encouragement from my family and friends.
Their words give me power in order to overcome all the embarrassment, discouragament and other difficultics. Thanks for everything helping ms to got this day. Hanoi, 23/03/2018 Nguyen Hang Phuong Figure 3-16: Comparison between SSCI's and Mormants using the purpose subband filters in Jaif transition: a) FL, b) F2, ¢)MI and 4) M2. - ee Figaro 3-17; Comparison hotweon SSCPs andForrnanis using the purpose subband fliers in /ae/ transition: 8) F1, b) F2, ¢) MI and đ) M2 - 45 Figure 3-18: Comparison between SSCFs and Fommants using the purpose subband filters in /ao/ transtior a) F1, b) F2, c) M1 and d) M2.
"- Figure 3-29: Comparison between SSCKs and Formants using the purpose subband filters in /au/ transition: a) Fl, b) F2, c) M1 and d) M2. Figaro 3-20: Compearisơm betwoow SSCBs andFormnants w Aa/ transition: a) 1/1, b) F2, €) MI and đ) M2 Figure 3-21: Comparison between SSCFs and Formants using the purpose subband filters in Jenf transition: ø) F1, b) F2, ¢) MI and d) M2. 30 Figure 3-22: Comparison between SSCFs and Formants using the purpose subband filters in Joa! transition: a) Fl, b) F2, e) M1 and đ) M2. ve Figure 3-23; Comparison between SSCFs and Formanis using the purpose subband Biles in ‘Aua/ transition a) E1, b} 1⁄2, ¢) M1 and d) M2 - 52 Figwe 3-24; Compasison between SSCFs and Fommants using the purpose subband filters in Any transition: a) Fl, b) F2, o) MI and d) M2 - 33 Figure 3-25: Comparison between SSCFs and Formants using the purpose “subband filters im Aui/ transition: a) Fl, b) F2, ø) MI and đ) M2 - eee | Figure 3-2 SSCT angles!2 in SSCP I/SSCP2 plane [13] - 56 Figure 3-27: SSC¥ angles calculated from the purpose definitionin /ai/ transition: a) K, b) T2, c) M1 and d) M2 - .39 Figure 3-28: SSCIF angles culeutated from the purpose doRnilion im /ao/ bamsilion: a) Fl.
hi) F2,c) MI and d) M2. woe 60 Figure 3-29; SSCF angles caleulated trom ‘the purpose p » definition irin /ao/ transition: a) Fl, b) F2, c) MI and 4) M2 ce 61 Figure 3-30: SSCF angles calonlated from the purpose definition in /aw transition: a) K, b) T2, ) MI and d) M2 - .-02 Pignre 3-31: SSCP angles calculaled fromm the purpose definition in Au/ transition: a) Ft, b) F2, 0) Mi and d) M2. sone 64 Figure 3-32: SSCF angles calculated fiom the purpose definitionin /ea/ transition. a) FI, b) 2, c) MI and đ) M2 eos Figure 3-33: SSCF angles calonlated from the purpose definition in foa/ transition: a) K, b) T2, ¿) MI and 4) M2 ce OF Tigure 3-34: SSCP angles calculated from the pirpose definition in /ua/ transition: a) Fl, bộ F2, c) MI and d) M2.
68 Rewe 3-35; SSCF angles calculated from the purpose definition in Aw transition: a} F1, b) 2,¢) MI and đ) M2. 70 Paw 3-36: SSCF angles calculated from the purpose definition in /ai/ transition: a) F1, b) T2, ©) MI and d) M2 - 72 Figure 4-1: A schematic overview of the Kaldi toolkit [30] 75 LIST OF FIGURES Figure 2-1: An outline of a typical speech recognition system [3| 13 Figure2-2, a) Singlo-tone sine wave of 10 Hz sampled using a sampling ficqueney of 1000 Hz, b) Magnitude spectrum of single-tone sine wave respectively [9] 15 Figure 2-3: a) Multitone sine wave of 10, 50 and 100 Hz sampled using a sampling frequency of 1000 Hz, b) Magnitude spactrum of corresponding multi-tonc sine wave [9] l6 Figure 2-4: a) Non-stationary multi-tone sins wavz of L0, 50 and 100 LIz sampled using a sampling frequency of 1000 Hz, b) Magnitude spectra of corresponding non-statiouary mmulti- tone sing wave [9] 7 Figure 2-5: a) Speech signal for the Hindi word “sakshaat”, b) Corresponcing spectra of different sepments of the Hindi speech signal |9|.- Figure2-6: Flow chart for MPCC computation [21] - ny) Figure 3-1: The algorithm is for extracting SSCs. 35 Figure 3-2: The shape of six-tuiangle overlapped subband filters for compuing SSCF.26 Figure 3-3: SSCF purancicrs extraction from a specch sigual [allowing frame by frarns [13] "- Figure 3-4: Compaison results between formants and SSCFs in /ai’ when apply six subband filters: a) F1, b) P2, 6) ML and d) M2 ce DT Pigure 3-5:'The shape of five-triangle overlapped subband filters for computing SSCF.28 Figure 3-6: Comparison results between formants and SSCFs in /ai/ when apply five overlap subband filters: a) Fi, b) F2, ø) MI and d) M2 - 20 Figure 3-7: The method for evaluation the effect of the number of subband filters on SSCF results. 230 Figure 3-8: Comparison when using 5 or 6 Triangular Subband Filters in /ai/ transition: a) Fl, b) F2, 6) ML and d) M2.
32 Figure 3-9: Comparison when using 5 or 6 Triangular Subband Fillers in /ac/ transition: #) F1, ) P2, ¢) MI and d) M2. 3⁄4 Figure 3-10: Comparison when using 5 or 6 Triangular Subband Filters in /ao/ transition: a) Fl, b) F2, o) Mi and d) M2. 36 Figare 3-1]: Comparison ‘when using Soré6 Triangutor Subband Filters in én tronsition a) the first female (F1), b) the second female (1:2), ¢) the first male (M1), d) the second male (2).38 Figure 3-12: [a] trajectories for native French speakers at mormal rate: a) FI-F2 plane at publishcation [25]; SSCF1-SSCF2 plane from meesurament results when b) using 6 triangular filters, ¢) using5 triangular filtcrs. 139 Figure 3-13: The definition of subband Gilter with equal Tangth in mncl-suale: a) ñve-bianglc subband filters, b) six triangle subband filters.
„41 Figure 3-14: The shape of new ptoposal six subband filters. 2A Figure 3-15: a) The trajectories in Vowel-to-Vowel French transition, 1 obtained with simulation [27], [28], Fuench vocalic triangle in SSCF1-SSCF2 plane: b) For two native females, c) For TWO THAIVE HH8 ÏC§,. chien 243 Figure 3-16: Comparison between SSCI's and Mormants using the purpose subband filters in Jaif transition: a) FL, b) F2, ¢)MI and 4) M2. - ee Figaro 3-17; Comparison hotweon SSCPs andForrnanis using the purpose subband fliers in /ae/ transition: 8) F1, b) F2, ¢) MI and đ) M2 - 45 Figure 3-18: Comparison between SSCFs and Fommants using the purpose subband filters in /ao/ transtior a) F1, b) F2, c) M1 and d) M2.
"- Figure 3-29: Comparison between SSCKs and Formants using the purpose subband filters in /au/ transition: a) Fl, b) F2, c) M1 and d) M2. Figaro 3-20: Compearisơm betwoow SSCBs andFormnants w Aa/ transition: a) 1/1, b) F2, €) MI and đ) M2 Figure 3-21: Comparison between SSCFs and Formants using the purpose subband filters in Jenf transition: ø) F1, b) F2, ¢) MI and d) M2. 30 Figure 3-22: Comparison between SSCFs and Formants using the purpose subband filters in Joa! transition: a) Fl, b) F2, e) M1 and đ) M2.