Dilu TTS (සිංහල) — `si_LK-dilu-medium`

Dilu is an open-source Sinhala (si_LK) text-to-speech voice for Piper. It is a medium-quality single-speaker female voice fine-tuned for natural conversational Sinhala.


Language	Sinhala (`si`)
Quality	Medium (22.05 kHz)
Format	Piper ONNX
Speakers	1
Base checkpoint	Hindi `hi_IN-rohan-medium` (`epoch=3190`)
Fine-tune	+400 epochs → `epoch=3589`

Audio samples

Sample	Text	Audio
Greeting	ආයුබෝවන්. මම ඩිලු.	01_ayubowan.wav
Weather	අද කාලගුණය හොඳයි.	samples/02_weather.wav
Beaches	ශ්‍රී ලංකාවේ ලස්සන වෙරළ බොහෝයි.	samples/03_beaches.wav
Call again	කරුණාකර තවත් වරක් කතා කරන්න.	samples/04_call_again.wav
Thanks	ස්තූතියි. ඔබට සුභ දවසක් වේවා.	samples/05_thank_you.wav

Quick start

Piper CLI

pip install piper-tts
# also install espeak-ng (e.g. brew install espeak-ng)

echo "ආයුබෝවන්, මම ඩිලු." | piper \
  --model onnx/si_LK-dilu-medium.onnx \
  --config onnx/si_LK-dilu-medium.onnx.json \
  --output_file dilu.wav

Python

from piper import PiperVoice
from pathlib import Path
import wave

voice = PiperVoice.load(
    "onnx/si_LK-dilu-medium.onnx",
    config_path="onnx/si_LK-dilu-medium.onnx.json",
)
with wave.open("dilu.wav", "wb") as wav_file:
    voice.synthesize_wav("ආයුබෝවන්, මම ඩිලු.", wav_file)

Model files

onnx/si_LK-dilu-medium.onnx       # inference weights (~61 MB)
onnx/si_LK-dilu-medium.onnx.json  # Piper config (phoneme map, sample rate, …)
samples/                          # example WAVs synthesized with this voice

Training notes

Dataset: female Sinhala LJSpeech-format clips (1,251 utterances)
Framework: Piper VITS / PyTorch Lightning
Fine-tuned from the public Hindi Rohan medium Piper checkpoint
Exported with piper_train.export_onnx

Intended use

Offline / on-device Sinhala TTS in apps, accessibility tools, and assistants
Research and education on low-resource TTS

Limitations

Single speaker; not multi-speaker or voice-cloning
Best on standard Sinhala orthography; rare phonemes / code-switching may sound uneven
Medium quality — not studio / neural-prosody SOTA

License

MIT — free to use, modify, and redistribute with attribution appreciated.

Citation

@misc{dilexus-dilu-tts,
  title        = {Dilu TTS: Open Sinhala Piper Voice (si_LK-dilu-medium)},
  author       = {Dilexus},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/chaturadilan/dilu-tts}},
  note         = {Fine-tuned Piper ONNX voice for Sinhala}
}

Acknowledgements

Piper by Rhasspy / Mike Hansen
Base voice: hi_IN-rohan-medium Piper checkpoint

Downloads last month: -; Downloads are not tracked for this model. How to track

Dilu TTS (සිංහල) — si_LK-dilu-medium