README.md

metadata

title: Khmer Text-to-Speech (ខ្មែរ TTS)
emoji: 🗣️
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 6.20.0
app_file: app.py
pinned: false
license: cc-by-nc-4.0
short_description: Natural Khmer speech from text — free demo, 2 voices
tags:
  - text-to-speech
  - tts
  - khmer
  - km
  - cambodia
  - f5-tts
  - audio
datasets:
  - DDD-Cambodia/khmer-speech-dataset

Khmer text-to-speech — ការបម្លែងអត្ថបទជាសំឡេងខ្មែរ

Type Khmer text, hear natural Khmer speech. Free demo, two voices (male/female).

វាយអត្ថបទខ្មែរ ហើយស្តាប់សំឡេងអានដោយ AI — ឥតគិតថ្លៃ។ អាចអានលេខ កាលបរិច្ឆេទ និងអត្ថបទវែងៗបានដោយស្វ័យប្រវត្តិ។

How it works: an F5-TTS model fine-tuned on ~1,000 hours of Khmer speech, with a purpose-built Khmer text front-end — automatic number verbalization (២៥ → ម្ភៃប្រាំ), sentence-aware chunking for paragraphs, and Khmer Unicode repair. In our held-out evaluation it transcribes more intelligibly (ASR CER) than Google Translate's Khmer voice and Meta MMS.

• Runs on ZeroGPU (free) — generations take a few seconds.
• Model weights are private; this Space serves audio only.
• Non-commercial research demo (base checkpoint is CC-BY-NC-4.0).
• Speech data by Digital Divide Data (CC-BY-SA-4.0) — thank you for making Khmer speech data exist.