Aurlo TTS Models

Pre-trained text-to-speech models for Aurlo - a privacy-first TTS desktop application.

Models

This repository contains the Kokoro-82M ONNX models optimized for fast, high-quality speech synthesis.

Available Files

File	Size	SHA256 Checksum	Description
kokoro-v1.0.onnx	~310 MB	7d5df8ecf7d4b1878015a32686053fd0eebe2bc377234608764cc0ef3636a6c5	Kokoro TTS model (ONNX format)
voices-v1.0.bin	~27 MB	bca610b8308e8d99f32e6fe4197e7ec01679264efed0cac9140fe9c29f1fbf7d	Voice embedding data (54 voices)

Voices Included

The voices-v1.0.bin file contains embeddings for 54 high-quality voices across multiple languages:

• American English - Multiple male and female voices
• British English - Various regional accents
• Japanese - Native speakers
• Mandarin Chinese - Multiple tones
• Spanish - International variations
• French - European French
• Hindi - Indian voices
• Italian - Native speakers
• Portuguese - Brazilian and European

Model Details

• Base Model: Kokoro-82M (87 million parameters)
• Architecture: ONNX Runtime optimized
• Performance: 210× realtime speed on modern hardware
• Sample Rate: 24kHz
• Format: 16-bit PCM WAV output
• License: Apache 2.0 (see LICENSE file)

Usage

These models are automatically downloaded by Aurlo on first launch. They are stored in the application's data directory:

• macOS: ~/Library/Application Support/com.aurlo.app/models/
• Windows: %APPDATA%\com.aurlo.app\models\

Manual Download

You can download the models manually from the Releases page.

Verification

Always verify the SHA256 checksums after downloading:

# macOS/Linux
shasum -a 256 kokoro-v1.0.onnx
shasum -a 256 voices-v1.0.bin

# Windows (PowerShell)
Get-FileHash kokoro-v1.0.onnx -Algorithm SHA256
Get-FileHash voices-v1.0.bin -Algorithm SHA256

Attribution

These models are based on the Kokoro-82M project and have been converted to ONNX format for optimal performance.

Original Model Credits:

• Kokoro-82M: hexgrad on Hugging Face
• ONNX Conversion: thewh1teagle/kokoro-onnx

License

These models are released under the Apache License 2.0 - see the LICENSE file for details.

The models maintain the same license as the original Kokoro project to ensure compatibility and proper attribution.

Version History

v1.0 (October 2025)

• Initial release
• Kokoro-82M base model
• 54 voice embeddings
• ONNX optimized for cross-platform deployment

Technical Specifications

• Input: Text (UTF-8 string)
• Output: 24kHz 16-bit mono PCM audio
• Inference Engine: ONNX Runtime
• Supported Platforms: macOS (ARM64, x86_64), Windows (x86_64)
• GPU Acceleration: Optional (CPU-only supported)

Support

For issues related to:

• Model performance: Open an issue in this repository
• Aurlo application: Open an issue in seed-blocks/aurlo
• Original Kokoro model: See hexgrad/Kokoro-82M

Citation

If you use these models in your research or application, please cite the original Kokoro project:

@misc{kokoro2024,
  title={Kokoro-82M: Compact Text-to-Speech Model},
  author={hexgrad},
  year={2024},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/hexgrad/Kokoro-82M}}
}

---

Repository: seed-blocks/aurlo-models Main Application: seed-blocks/aurlo License: Apache 2.0