piper-voices

🎙️ Piper Neural TTS Voices Mirror (Sherpa-ONNX Bundles)

High-speed global CDN mirror of 129 curated single-speaker Piper VITS neural TTS models packaged for zero-config, offline, on-device text-to-speech synthesis in sherpa-onnx and the Anuwad cross-platform application.

---

📌 Features & Highlights

• 100% Single-Speaker Studio Models: Every model in this repository is configured for sid: 0 out-of-the-box — no speaker ID lookup, multi-speaker indexing, or complex configuration required.
• Self-Contained Model Bundles: Each .tar.bz2 archive contains everything needed for fully offline text-to-speech inference:
  • <model>.onnx — The optimized neural acoustic model (VITS architecture)
  • tokens.txt — Phoneme vocabulary and symbol mappings
  • espeak-ng-data/ — Complete phonemization directory for offline text processing
• High-Speed CDN Acceleration: Hosted directly on Hugging Face's global edge infrastructure powered by Cloudflare, enabling fast, resilient model downloads worldwide.
• Broad Language Coverage: 129 voices spanning 42+ languages and regional accents across Europe, Asia, and the Americas.

---

🔗 Download URL Format

Any voice archive can be downloaded directly from the Cloudflare CDN edge via:

https://huggingface.co/4-alokk/piper-voices/resolve/main/<voice_id>.tar.bz2

Example (cURL):

curl -L -O https://huggingface.co/4-alokk/piper-voices/resolve/main/vits-piper-en_US-lessac-medium.tar.bz2

Example (Python with huggingface_hub):

from huggingface_hub import hf_hub_download

# Download any Piper bundle with local caching
bundle_path = hf_hub_download(
    repo_id="4-alokk/piper-voices",
    filename="vits-piper-hi_IN-pratham-medium.tar.bz2",
    repo_type="model",
)
print(f"Model downloaded to: {bundle_path}")

Example (Flutter / Dart with dio):

final url = 'https://huggingface.co/4-alokk/piper-voices/resolve/main/$voiceId.tar.bz2';
await dio.download(url, savePath);

---

👥 Credits, Contributors & Acknowledgments

This repository stands on the shoulders of remarkable open-source speech synthesis pioneers and dataset creators. Deepest gratitude to:

• Michael Hansen (@synesthesiam) and the Piper TTS Community:
  • For designing and training the fast, high-quality VITS-based neural text-to-speech engine.
• Fangjun Kuang (@csukuangfj) and the k2-fsa / sherpa-onnx project:
  • For converting, quantizing, packaging, and optimizing Piper models for embedded and cross-platform runtime execution.
• Voice Dataset Creators & Community Contributors:
  • Thorsten Müller (Thorsten Voice) for top-tier open German voice datasets.
  • Keith Ito and the LJSpeech corpus team.
  • LibriTTS, M-AILABS, and OpenSLR speech contributors across all 42+ languages.
  • Community voice actors who generously contributed high-fidelity audio recordings.
• Curated, Mirrored & Maintained by:
  • Alok Kumar (@4-alokk) for the Anuwad open communication and translation ecosystem.

---

📋 Catalog of Available Voices (129 Models)