agentvibes/piper-custom-voices

AgentVibes Custom Piper Voices

High-quality custom Piper TTS voices for the AgentVibes project.

Voices Included

1. Kristin

• Type: US English female
• License: Public Domain
• Quality: Medium (2000 epochs)
• Dataset: ~11.5 hours from LibriVox.org
• Creator: Bryce Beattie
• Source: https://brycebeattie.com/files/tts/

2. Jenny

• Type: UK English female (Irish accent)
• License: CC BY (Attribution required)
• Quality: High
• Dataset: Dioco (Kaggle)
• Creator: Bryce Beattie
• Source: https://brycebeattie.com/files/tts/

3. Tracy (ManyVoice)

• Type: Multi-speaker (16 voices: 12 US English, 4 UK English)
• License: Public Domain
• Quality: Medium (400 epochs)
• Dataset: ~8 hours per voice from LibriVox.org
• Creator: Bryce Beattie
• Source: https://brycebeattie.com/files/tts/

Usage

With AgentVibes

These voices are automatically available when using AgentVibes with Piper TTS provider.

Install AgentVibes:

npx agentvibes@beta install

Switch to Piper provider:

/agent-vibes:provider switch piper

Use custom voices:

/agent-vibes:switch kristin
/agent-vibes:switch jenny
/agent-vibes:switch tracy

Manual Download

# Download all voices
git clone https://huggingface.co/agentvibes/piper-custom-voices

# Or download individual voices
wget https://huggingface.co/agentvibes/piper-custom-voices/resolve/main/kristin.onnx
wget https://huggingface.co/agentvibes/piper-custom-voices/resolve/main/kristin.onnx.json

With Piper TTS

echo "Hello from Kristin" | piper \
  --model kristin.onnx \
  --output_file output.wav

File Sizes

• kristin.onnx: ~64MB
• jenny.onnx: ~64MB
• tracy.onnx: ~74MB

Attribution

All voices created by Bryce Beattie (https://brycebeattie.com/files/tts/)

Creator's Permission

"Feel free to use these for any legal and ethical purpose. If somebody wants to upload these to HuggingFace or somewhere similar, you have my blessing."

— Bryce Beattie, https://brycebeattie.com/files/tts/

Voice Licenses

• Kristin: Public Domain (LibriVox recordings)
• Jenny: CC BY (Attribution required - Dioco dataset)
• Tracy: Public Domain (LibriVox recordings)

License

• Repository License: Apache 2.0
• Voice Licenses: See individual voice descriptions above
• Dataset Sources: LibriVox.org (Public Domain), Dioco/Kaggle (CC BY)

Links

• AgentVibes: https://github.com/paulpreibisch/AgentVibes
• Piper TTS: https://github.com/rhasspy/piper
• Voice Creator: https://brycebeattie.com/files/tts/
• HuggingFace Voices: https://huggingface.co/rhasspy/piper-voices

Technical Details

• Format: ONNX (Open Neural Network Exchange)
• Engine: Piper TTS (https://github.com/rhasspy/piper)
• Training: LibriVox public domain audiobooks
• Quality: Medium to High (400-2000 epochs)
• Enhancements: Sound quality improvements applied during training

Contributing

Found a bug or want to suggest a voice? Open an issue at: https://github.com/paulpreibisch/AgentVibes/issues