v0.1.2

v0.1.2 (2026-09-29)

Bug Fixes

• Deploy TinyTalk from a writable application directory (#9, 17d6fe8)

Continuous Integration

• Automate release preparation and publishing (#9, 17d6fe8)

• Use Python Semantic Release (#9, 17d6fe8)

• Validate pull request titles (#9, 17d6fe8)

Documentation

• Document backend model licenses and Nix setup (#9, 17d6fe8)

• Link upstream model license sources (#9, 17d6fe8)

Refactoring

• Remove Nix runtime path indirection (#9, 17d6fe8)

• Remove obsolete Python environment path (#9, 17d6fe8)

• Simplify Nix deployment (#9, 17d6fe8)

• Simplify Nix runtime setup (#9, 17d6fe8)

• Trim Nix runtime plumbing (#9, 17d6fe8)

Testing

• Exercise Nix upgrade lifecycle (#9, 17d6fe8)

---

Detailed Changes: v0.1.1...v0.1.2

v0.1.1

Short update to tinytauk engine which reduces cost to regenerate bad chunks.

Full Changelog: v0.1.0...v0.1.1

Initial Release

TinyTalk v0.1.0

First release of TinyTalk, an OpenAI-compatible TTS server for modest machines.

Features

• Three synthesis backends: NeuTTS, OmniVoice, and TinyTAuK.
• OpenAI-compatible /v1/audio/speech endpoint.
• OmniVoice automatic voice generation, voice design, and reference cloning.
• TinyTAuK v0.2.0 integration with configurable runtime profiles.
• Shared audio chunking, stitching, and quality evaluation.
• NixOS module for declarative deployment.
• Unified, locked Python 3.13 runtime using uv.
• Torch/Torchaudio 2.11 and Transformers 5.17.

Known limitations

• Named voice selection through the voice field is not yet implemented.
• Only one backend is active per service instance.
• The checked-in Python runtime uses CPU PyTorch wheels.