I am a Senior ML Engineer at the Digital Government Committee at the moment, previously I led the Front-end development team there. My interests are product design, product development, graphics programming, music and deep learning.
Products
Verify.gov.kh
A national document verification platform for Cambodia
StopCOVID.gov.kh
An innovative national COVID-19 tracing web app
Sarika.gov.kh
High-quality Khmer Text-to-speech engine for government
Joon.com.kh
Local food directory and food review social network
Koh Santepheap Daily
A well-known newspaper company. I work on the native Android client and its backend API
Pi Pay (Android)
The most innovative and complete cashless payment platform in Cambodia
Khmer Dictionary
The blazingly fast Khmer dictionary
KhmerOCR
An OCR app for any languages
CarnetDia (Android)
The first mobile app that raises awareness about diabetes in Khmer language
QR.GOV.KH
A productive and fast QR generator
Companies I worked with
Research
Khmer Text-to-Speech System Proprietary
An end-to-end system for Khmer speech synthesis which includes a custom text phonemizer, text normalizer, tokenizer and vocoder optimized for the Khmer language.
A Fast Khmer Optical Character Recognition (Document only). The model was trained on 3M text lines with over 800+ Khmer fonts.
Inverse Text Normalization for Khmer Automatic Speech Recognition
A fast Khmer Forced Aligner powered by Wav2Vec2CTC and Phonetisaurus
A Khmer pronounciation toolkit
A Python library for Khmer Text Normalization and Verbalization Toolkit
Text to Audio forced aligner similar to KFA but it was trained with Montreal Forced Aligner instead of Wav2Vec
Punctuation Restoration for Khmer language
Training from scratch using Masked Language Modeling task on 5M Khmer sentences or 162M words or 578K unique words for 1M steps.
Open-Source
A declarative Canvas layout engine for JavaScript with advanced rich text support.
Python binding for Phonetisaurus
An open-source part of speech tagger for Khmer language using BiLSTM.
Khmer encoding normalization implementation in C++.
A (fast) Khmer word segmentation toolkit
Khmer Unicode Converter for JavaScript
A multi-processing audio check
A Khmer word segmentation tool built for NIPTICT (now CADT) Khmer Word Segmentation CRF model.
Python binding for CRF++
A Fast Speech Enhancement toolkit using Conv-TasNet (Yi Luo, Nima Mesgarani)
A CRFSuite port for Node, Browser & Deno
A simple and lightweight Khmer lunar calendar. (1.7kB minified)
Convert Android VectorDrawable to SVG
An efficient encryption for JSON binary
Presentations
Beliefs
Great products are built with focus, effort and attention-to-detail.
Finishing a project is way harder than starting a new one.
Nothing can beat practice.
Anger destroys you and people around you.
The best is the enemy of the good.
Body should operate without substances.
Complexity is often solved by brute-force persistence.
Contact
You can reach me at seanghay.dev@gmail.com