All projects
● ML 2026

Khmer Speech-to-Text

Fine-tuned Whisper for Khmer speech.

Role
ML engineer
Context
AI course project · RUPP
Stack
PyTorch · Hugging Face · Gradio

Speech recognition for Khmer lags far behind major languages. In this project I fine-tuned OpenAI’s Whisper model on Khmer speech so it can transcribe spoken Khmer into text, and published an interactive demo.

What I did

  1. 01

    Prepared Khmer speech recordings and their transcripts for training.

  2. 02

    Fine-tuned the Whisper Small model on the Khmer data over several iterations.

  3. 03

    Built a Gradio web demo where anyone can record or upload audio and get a transcript.

  4. 04

    Hosted the demo on Hugging Face Spaces.

Highlights

Speech to text

Transcribes spoken Khmer.

Fine-tuned Whisper

Adapted to Khmer speech.

Live demo

Runs in the browser on Hugging Face.

Built with

PyTorchHugging FaceGradioPython
Next projectKhmer Word SegmentationSplits space-less Khmer text into words. All projects