Lao TTS
Lao TTS
Lao TTS
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
<strong>Lao</strong> Text-to-speech synthesis<br />
HMM- based Method<br />
National Authority for Science Technology<br />
(NAST) , <strong>Lao</strong> PDR<br />
National Electronic and Computer Technology Centre<br />
(NECTEC), Thailand<br />
12/20/08 1
Status<br />
<br />
<br />
<br />
<br />
<br />
Phone inventory design for <strong>TTS</strong> (Done)<br />
Preparing resources and tools (Done)<br />
Designing and creating <strong>Lao</strong> text processing tools (Done)<br />
Sentence selection & speech corpus preparation (Done)<br />
Training and Creating synthesizer by HMM-based speech<br />
synthesis system toolkit (being trained)<br />
12/20/08
Overview<br />
<strong>Lao</strong> language and <strong>Lao</strong> writing system<br />
Phonemes design for <strong>Lao</strong> <strong>TTS</strong><br />
<strong>Lao</strong> Sound systems<br />
Syllable structure and syllable breaking<br />
<strong>Lao</strong> <strong>TTS</strong> Development process<br />
12/20/08 3
Introduction to <strong>Lao</strong> language<br />
<strong>Lao</strong> language is the official language of <strong>Lao</strong>s. It is a tonal<br />
Language of Tai family an is closely related to the Isan<br />
Language of the northeast region of Thailand<br />
<strong>Lao</strong> spoken Language can be divided into 3 main groups<br />
<br />
<br />
<br />
Vientiane <strong>Lao</strong><br />
Northern <strong>Lao</strong> (Luang Prabang)<br />
Southern <strong>Lao</strong> (Champasak)<br />
12/20/08 4
Writing systems<br />
<br />
<br />
<br />
<br />
<strong>Lao</strong> language are write from left to right<br />
No space between word and sentence<br />
One word can be one or more syllables<br />
<strong>Lao</strong> alphabets: <strong>Lao</strong> language are contained 78 characters, divided<br />
in to 5 groups such as:<br />
<br />
<br />
<br />
<br />
<br />
33 consonants<br />
28 vowels<br />
4 tone mark<br />
10 <strong>Lao</strong> digits<br />
3 special characters<br />
12/20/08 5
Writing system Cont.<br />
Consonants:<br />
<br />
The single consonants can all be used as the main consonant of<br />
a syllable and somes can be used at the end of a syllable or use<br />
as the consonantal.<br />
Use as final consonant<br />
Common Cluster consonants<br />
12/20/08 6
Writing system Cont.<br />
<br />
Mixed Consonants<br />
12/20/08 7
Writing system Cont.<br />
<br />
Vowels: we separated into 2 group<br />
Single vowels and mix vowels as you see the first<br />
line of image bellow is a single vowels, last two line<br />
is mix vowels.<br />
12/20/08 8
Writing system Cont.<br />
1. Tone marks<br />
Tonal marks are another group of alphabet letters which are<br />
symbols characterizing the change of the sounds of a<br />
syllable to follow the rule of consonants sounds with short,<br />
medium, low, high tone.<br />
12/20/08 9
1. Special symbol<br />
Writing system Cont.<br />
2. <strong>Lao</strong> digits<br />
12/20/08 10
Phoneme design for <strong>Lao</strong> <strong>TTS</strong><br />
See All<br />
12/20/08 11
1. Consonants sound<br />
Sound systems<br />
High consonants: kh,s,th,ph,f,h (ຂ ສ ຖ ຜ ຝ ຫ)<br />
Mid consonants: k,c,d,t,b,p,j,z (ກ ຈ ດ ຕ ບ ປ ຢ ອ)<br />
Low consonants: kh,ng,s,y,th,n,ph,f,m,l,w,h (ຄ ງ ຊ ຍ ທ ນ ພ<br />
ຟ ມ ລ ວ ຮ)<br />
2. Tones sound<br />
12/20/08 12
3. vowels sound<br />
Sound systems<br />
aa, ii, vv, uu, qq, ee, xx, oo, @@, uua, iia, vva<br />
Long<br />
Short<br />
(ອາ<br />
ອີ ອື ອູ ເອີ ເອ ແອ ໂອ ອໍ ອົວ ເອຍ ເອືອ)<br />
a, i, v, u, q, e, x, o, @, ua, ia, va<br />
(ອະ<br />
ອິ ອຶ ອຸ ເອິ ເອະ ແອະ ໂອະ ເອາະ ອົວະ ເອັຍ<br />
ເອຶອ)<br />
12/20/08 13
12/20/08<br />
Sound systems
final sound or tone<br />
mark <br />
the first consonants of<br />
word <br />
final sound of word<br />
are<br />
unstoped:th<br />
e word that<br />
ends with<br />
nasal<br />
sound:<br />
m,n,ng (ມ,ນ<br />
,ງ) or<br />
unstoped<br />
vowels<br />
sound.<br />
final sound of word<br />
is stoped<br />
:k,ng,d,b (ກ,ງ<br />
,ດ,ບ),<br />
and<br />
vowels length<br />
is sorth<br />
a,i,v,u,q,e,x,o,<br />
@,ua,ia,va<br />
(ະ ິ ຶ<br />
ຸ ເອິ ເອະ<br />
ແອະ ໂອະ ເອາະ<br />
ອົວະ ເອັຍ<br />
ເອຶອ) (the<br />
word ends<br />
with stoped<br />
consonant or<br />
vowel sound)<br />
final sound of word is<br />
stoped:k,ng,d,b (ກ,ງ<br />
,ດ,ບ),<br />
and vowels<br />
length is long:<br />
aa,ii,vv,uu,qq,ee,xx,o<br />
o,@@,uua,iia,vva (າ<br />
ີ ື ູ ເອີ ເອ<br />
ແອ ໂອ ອໍ ອົວ ເອຍ<br />
ເອືອ) (the word ends<br />
with stoped<br />
consonant there are<br />
no long,stoped<br />
vowels)<br />
any word with the<br />
first tone<br />
mark (x່)<br />
any word with the second<br />
tone mark (x້)<br />
high consonants:<br />
kh,s,th,ph,f,h (ຂ<br />
ສ ຖ ຜ ຝ ຫ)<br />
rising (4) high (3) low falling (0) mid (1) high falling (2)<br />
mid consonants:<br />
k,c,d,t,b,p,j,z (ກ<br />
ຈ ດ ຕ ບ ປ ຢ ອ)<br />
rising (4) high (3) low falling (0) mid (1) high falling (2)<br />
low consonants:<br />
kh,ng,s,y,th,n,ph,<br />
f,m,l,w,h (ຄ ງ ຊ<br />
ຍ ທ ນ ພ ຟ ມ ລ<br />
ວ ຮ)<br />
high (3) mid (1) high falling (2) mid (1) high falling (2)<br />
12/20/08
<strong>Lao</strong> Syllable structure<br />
<strong>Lao</strong> syllable was designed as “CV, CVC, CVV,<br />
and CVVC ”<br />
- C: Main or nuclear Consonants<br />
- V: vowels<br />
- C: Consonantal or final consonants<br />
12/20/08 16
<strong>Lao</strong> Syllable breaking<br />
12/20<br />
20/0808 17
<strong>Lao</strong> <strong>TTS</strong> development<br />
Have 2 main modules: Natural Language Processing or NLP<br />
Module and Digital signal processing or DSP modules<br />
Text<br />
Natural Language<br />
Processing (NLP)<br />
Phone<br />
Transcription<br />
Prosody<br />
Digital Signal<br />
Processing (DSP)<br />
Speech<br />
12/20/08 18
Natural Language Processing<br />
Text analysis<br />
Implemented using <strong>Lao</strong> text corpus (5 MB)<br />
Sentences end marker by space and sentence long<br />
(11,159 sentences)<br />
Syllable breaking based on <strong>Lao</strong> syllabification<br />
techniques<br />
Grapheme-to-phoneme (G2P) was implemented<br />
using Finite state Machines (FSMs) Toolkit<br />
12/20<br />
20/0808
12/20<br />
20/0808<br />
Speech corpus<br />
Natural Language Processing<br />
1,619 sentences were selected for recording<br />
Including 60 phoneme and 5 tones<br />
By female speaker<br />
Time used: 15 hours<br />
Sentence cutting by Manual<br />
Prosody generation<br />
Using HTS toolkit to generate speech parameter<br />
Mel-Cestrum (MCEP), duration and Log fundamental<br />
frequency (Log F0) were extracted from each utterance in<br />
the speech corpus
Digital Signal Processing (DSP)<br />
HMM-Based for speech synthesizer and now under<br />
Labeling files preparation and being training<br />
12/20/08
HTS Labeling files preparation<br />
mono.dic : list all lao syllables with phones and tones<br />
in speech database<br />
word.mlf : list all lao syllable with sentence marker in<br />
speech database<br />
mono.unit: list all phones with tone in speech<br />
database<br />
mono.list : list of all phonemes in speech database.<br />
questions_qst<strong>Lao</strong>001.hed : list of all context and<br />
properties format for tree-based context clustering.<br />
12/20/08
mono.dic<br />
word.mlf<br />
12/20<br />
20/0808
12/20<br />
20/0808<br />
questions_qst<strong>Lao</strong>001.hed
Thank you for your attention<br />
Khop cai lai lai !<br />
12/20/08 25