ੴ Punjabi Neural TTS Voices (ਪੰਜਾਬੀ ਆਵਾਜ਼ਾਂ)
High-Fidelity Punjabi Text-to-Speech (VITS ONNX)
Published by Nam-toon Studio & Gurpreet Singh as part of sovereign Indic speech AI research under AMRIT Research OS.
🌟 Model Overview
This repository hosts high-speed, edge-deployable VITS ONNX neural speech synthesis models for Punjabi (Gurmukhi). Built for sub-50ms latency on Raspberry Pi, Apple Silicon, and edge devices.
🎙️ Supported Voices & Dataset Linkage
Trained and aligned on the Nam-toon-studio/Punjabi-MultiSpeaker-Studio-TTS-Corpus (145 clips, 22.8 minutes of 22,050 Hz high-resolution studio speech):
spk1_male_orator: Conversational / Orator Voice (Bhai Ranjit Singh)spk2_male_literary: Classical Literary & Poetic Voice (Dr. Surjit Patar)spk3_female_broadcaster: Professional Broadcast Voice (BBC Punjabi Female)
📦 Model Files
pa-guru-female.onnx(114 MB): High-quality VITS neural generator.pa-guru-female.onnx.json: Model configuration, sample rate (22,050 Hz), and speaker mapping.pa-guru-multispeaker.onnx(114 MB): Multi-speaker model checkpoint.pa-guru-multispeaker.onnx.json: Multi-speaker metadata.
🚀 Quickstart with Python ONNXRuntime
import onnxruntime as ort
from huggingface_hub import hf_hub_download
# Download model & config
model_path = hf_hub_download('Nam-toon-studio/punjabi-tts-voices', 'pa-guru-female.onnx')
session = ort.InferenceSession(model_path)
print('Model inputs:', [i.name for i in session.get_inputs()])
print('Model outputs:', [o.name for o in session.get_outputs()])
📜 Citation
@misc{namtoon_punjabi_tts_2026,
author = {Gurpreet Singh (Nam-toon Studio)},
title = {Sovereign Punjabi Neural Text-to-Speech Voices},
year = {2026},
publisher = {Hugging Face},
howpublished = {\url{https://huggingface.co/Nam-toon-studio/punjabi-tts-voices}}
}