Lao TTS

Lao Text-to-speech synthesis
HMM- based Method
National Authority for Science Technology
(NAST) , Lao PDR
National Electronic and Computer Technology Centre
(NECTEC), Thailand
12/20/08 1

Status
Phone inventory design for TTS (Done)
Preparing resources and tools (Done)
Designing and creating Lao text processing tools (Done)
Sentence selection & speech corpus preparation (Done)
Training and Creating synthesizer by HMM-based speech
synthesis system toolkit (being trained)
12/20/08

Overview
Lao language and Lao writing system
Phonemes design for Lao TTS
Lao Sound systems
Syllable structure and syllable breaking
Lao TTS Development process
12/20/08 3

Introduction to Lao language
Lao language is the official language of Laos. It is a tonal
Language of Tai family an is closely related to the Isan
Language of the northeast region of Thailand
Lao spoken Language can be divided into 3 main groups
Vientiane Lao
Northern Lao (Luang Prabang)
Southern Lao (Champasak)
12/20/08 4

Writing systems
Lao language are write from left to right
No space between word and sentence
One word can be one or more syllables
Lao alphabets: Lao language are contained 78 characters, divided
in to 5 groups such as:
33 consonants
28 vowels
4 tone mark
10 Lao digits
3 special characters
12/20/08 5

Writing system Cont.
Consonants:
The single consonants can all be used as the main consonant of
a syllable and somes can be used at the end of a syllable or use
as the consonantal.
Use as final consonant
Common Cluster consonants
12/20/08 6

Writing system Cont.
Mixed Consonants
12/20/08 7

Writing system Cont.
Vowels: we separated into 2 group
Single vowels and mix vowels as you see the first
line of image bellow is a single vowels, last two line
is mix vowels.
12/20/08 8

Writing system Cont.
1. Tone marks
Tonal marks are another group of alphabet letters which are
symbols characterizing the change of the sounds of a
syllable to follow the rule of consonants sounds with short,
medium, low, high tone.
12/20/08 9

1. Special symbol
Writing system Cont.
2. Lao digits
12/20/08 10

Phoneme design for Lao TTS
See All
12/20/08 11

1. Consonants sound
Sound systems
High consonants: kh,s,th,ph,f,h (ຂ ສ ຖ ຜ ຝ ຫ)
Mid consonants: k,c,d,t,b,p,j,z (ກ ຈ ດ ຕ ບ ປ ຢ ອ)
Low consonants: kh,ng,s,y,th,n,ph,f,m,l,w,h (ຄ ງ ຊ ຍ ທ ນ ພ
ຟ ມ ລ ວ ຮ)
2. Tones sound
12/20/08 12

3. vowels sound
Sound systems
aa, ii, vv, uu, qq, ee, xx, oo, @@, uua, iia, vva
Long
Short
(ອາ
ອີ ອື ອູ ເອີ ເອ ແອ ໂອ ອໍ ອົວ ເອຍ ເອືອ)
a, i, v, u, q, e, x, o, @, ua, ia, va
(ອະ
ອິ ອຶ ອຸ ເອິ ເອະ ແອະ ໂອະ ເອາະ ອົວະ ເອັຍ
ເອຶອ)
12/20/08 13

12/20/08
Sound systems

final sound or tone
mark
the first consonants of
word
final sound of word
are
unstoped:th
e word that
ends with
nasal
sound:
m,n,ng (ມ,ນ
,ງ) or
unstoped
vowels
sound.
final sound of word
is stoped
:k,ng,d,b (ກ,ງ
,ດ,ບ),
and
vowels length
is sorth
a,i,v,u,q,e,x,o,
@,ua,ia,va
(ະ ິ ຶ
ຸ ເອິ ເອະ
ແອະ ໂອະ ເອາະ
ອົວະ ເອັຍ
ເອຶອ) (the
word ends
with stoped
consonant or
vowel sound)
final sound of word is
stoped:k,ng,d,b (ກ,ງ
,ດ,ບ),
and vowels
length is long:
aa,ii,vv,uu,qq,ee,xx,o
o,@@,uua,iia,vva (າ
ີ ື ູ ເອີ ເອ
ແອ ໂອ ອໍ ອົວ ເອຍ
ເອືອ) (the word ends
with stoped
consonant there are
no long,stoped
vowels)
any word with the
first tone
mark (x່)
any word with the second
tone mark (x້)
high consonants:
kh,s,th,ph,f,h (ຂ
ສ ຖ ຜ ຝ ຫ)
rising (4) high (3) low falling (0) mid (1) high falling (2)
mid consonants:
k,c,d,t,b,p,j,z (ກ
ຈ ດ ຕ ບ ປ ຢ ອ)
rising (4) high (3) low falling (0) mid (1) high falling (2)
low consonants:
kh,ng,s,y,th,n,ph,
f,m,l,w,h (ຄ ງ ຊ
ຍ ທ ນ ພ ຟ ມ ລ
ວ ຮ)
high (3) mid (1) high falling (2) mid (1) high falling (2)
12/20/08

Lao Syllable structure
Lao syllable was designed as “CV, CVC, CVV,
and CVVC ”
- C: Main or nuclear Consonants
- V: vowels
- C: Consonantal or final consonants
12/20/08 16

Lao Syllable breaking
12/20
20/0808 17

Lao TTS development
Have 2 main modules: Natural Language Processing or NLP
Module and Digital signal processing or DSP modules
Text
Natural Language
Processing (NLP)
Phone
Transcription
Prosody
Digital Signal
Processing (DSP)
Speech
12/20/08 18

Natural Language Processing
Text analysis
Implemented using Lao text corpus (5 MB)
Sentences end marker by space and sentence long
(11,159 sentences)
Syllable breaking based on Lao syllabification
techniques
Grapheme-to-phoneme (G2P) was implemented
using Finite state Machines (FSMs) Toolkit
12/20
20/0808

12/20
20/0808
Speech corpus
Natural Language Processing
1,619 sentences were selected for recording
Including 60 phoneme and 5 tones
By female speaker
Time used: 15 hours
Sentence cutting by Manual
Prosody generation
Using HTS toolkit to generate speech parameter
Mel-Cestrum (MCEP), duration and Log fundamental
frequency (Log F0) were extracted from each utterance in
the speech corpus

Digital Signal Processing (DSP)
HMM-Based for speech synthesizer and now under
Labeling files preparation and being training
12/20/08

HTS Labeling files preparation
mono.dic : list all lao syllables with phones and tones
in speech database
word.mlf : list all lao syllable with sentence marker in
speech database
mono.unit: list all phones with tone in speech
database
mono.list : list of all phonemes in speech database.
questions_qstLao001.hed : list of all context and
properties format for tree-based context clustering.
12/20/08

mono.dic
word.mlf
12/20
20/0808

12/20
20/0808
questions_qstLao001.hed

Thank you for your attention
Khop cai lai lai !
12/20/08 25