intellisr/sinhala-tts-piper-v1

Sinhala TTS Piper Version 1

Sinhala neural text-to-speech voice in the Piper format (ONNX + JSON config). The checkpoint was trained by Intellisr on a custom dataset produced by Access to Success Organization, using recordings of Mr Ashoka Weerawardhana. The model targets Sinhala (si): phonemization follows espeak-ng voice si, audio is 16 kHz, single speaker.

Credits and data provenance

Role	Organization / person
Training	Intellisr
Dataset	Access to Success Organization (custom Sinhala corpus)
Speaker voice	Mr Ashoka Weerawardhana

Use of this model should respect the dataset license and any agreements between the organizations and the speaker. The technical runtime remains subject to Piper’s GPL-3.0 toolchain when combined with upstream Piper components.

Model files

File	Description
model.onnx	Piper VITS-style ONNX graph for inference
model.onnx.json	Voice metadata, phoneme map, and default synthesis controls

Bundle metadata reports piper_version 1.0.0, num_speakers 1, phoneme_type espeak.

Intended use

Local/offline Sinhala speech synthesis with the Piper engine (CLI, Python piper-tts, or integrations such as Home Assistant), subject to your rights to use the voice and dataset.

How to run (Piper)

Install Piper (piper-tts) and ensure espeak-ng is available on your system (same stack as training).

echo "ආයුබෝවන්" | piper \
  --model model.onnx \
  --config model.onnx.json \
  --output_file sinhala.wav

Default inference settings from the config: noise_scale 0.667, length_scale 1.0, noise_w 0.8.

Training stack

Training used OHF-Voice/piper1-gpl. Environment included:

• requirements.txt from the Piper training repo
• cython>=0.29.0, piper-phonemize==1.1.0, librosa>=0.9.2, numpy>=1.19.0, onnxruntime>=1.11.0
• pytorch-lightning==1.7.0, torch==1.11.0, torchtext==0.12.0, torchvision==0.12.0
• torchaudio==0.11.0, torchmetrics==0.11.4 (compatibility pin)
• Monotonic align built via build_monotonic_align.sh
• System: espeak-ng (apt-get install espeak-ng on Debian/Ubuntu-style images)

License

This voice bundle is distributed under the GNU General Public License v3.0 to align with piper1-gpl (GPL-3.0). Review COPYING in the upstream repository if you redistribute combined works. Dataset and voice rights may impose additional terms beyond this README; obtain clarity from Intellisr and Access to Success Organization where needed.

Citation

Credit the speaker, dataset provider, trainer, and Piper as appropriate.

@misc{sinhala_piper_v1_intellisr,
  title = {Sinhala {TTS} {Piper} Version 1},
  author = {Weerawardhana, Ashoka and {Access to Success Organization} and {Intellisr}},
  year = {2025},
  note = {Piper ONNX voice; trained by Intellisr on Access to Success Organization custom dataset; Piper engine: OHF-Voice/piper1-gpl}
}

@software{piper2024,
  title = {Piper: Fast and local neural text-to-speech},
  url = {https://github.com/OHF-Voice/piper1-gpl},
  note = {Open Home Foundation}
}