Khmer TTS
Click here to load reader
Transcript of Khmer TTS
![Page 1: Khmer TTS](https://reader037.fdocuments.us/reader037/viewer/2022100415/555a409dd8b42ae1398b4ea1/html5/thumbnails/1.jpg)
Institut de Technologie du Cambodge (ITC)Génie Informatique et Communication (GIC)
TTS (Text-To-Speech)
Seangmeng LONG[[email protected]]
BarCamp
![Page 2: Khmer TTS](https://reader037.fdocuments.us/reader037/viewer/2022100415/555a409dd8b42ae1398b4ea1/html5/thumbnails/2.jpg)
22
What is TTS?
TTS stands for Text-To-Speech
It is a system (module) which takes as input text
in Khmer Unicode and produces Khmer speech
Electronic documents TTS system Khmer Speech
Input Output
![Page 3: Khmer TTS](https://reader037.fdocuments.us/reader037/viewer/2022100415/555a409dd8b42ae1398b4ea1/html5/thumbnails/3.jpg)
3
Our Method
Concatenation-Based Synthesis using Diphone
![Page 4: Khmer TTS](https://reader037.fdocuments.us/reader037/viewer/2022100415/555a409dd8b42ae1398b4ea1/html5/thumbnails/4.jpg)
4
Our Method (steps) Word Segmentation: → → →→→→→ →→ →→→→សម្�ចថាខ្មៅ�� សម្�ចថាខ្មៅ��→ , →→ →→→→សម្�ចថាខ្មៅ��
Text Normalization: ១២៤ ម្�យរយម្�បួ�ន→ , ម្�យពី�របួ�ន Text To Sound Conversion: → → kakthen
Syllabification: sa:la: → sa: . la:
Stress Assignment: sa: . la: → sə . la:
Sound Change: ចាក់� cak → caʔ
Intonation
Diphone Database
Integration
Applications Development (mail reader, doc reader, ...)
![Page 5: Khmer TTS](https://reader037.fdocuments.us/reader037/viewer/2022100415/555a409dd8b42ae1398b4ea1/html5/thumbnails/5.jpg)
5
New Statistical System Speech corpus
~450 sentences (~30 minutes) Automatic labeling
EHMM labeler Sphinx
Statistical parameter synthesis More natural, but buzzy
Unit selection Units of variable size (smallest unit is phone) More natural, but bad quality at join points
![Page 6: Khmer TTS](https://reader037.fdocuments.us/reader037/viewer/2022100415/555a409dd8b42ae1398b4ea1/html5/thumbnails/6.jpg)
6
Thanks for your attention.