v0.3.0

Added

• Version metadata endpoint: GET /version
• Adapter discovery endpoint: GET /adapters
• Adapter status endpoint: GET /adapters/{adapter_id}/status
• Streaming synthesis endpoint: POST /synthesize/stream (WAV streaming transport)

API Contract Updates

• OpenAPI spec regenerated and committed with the new endpoints.
• Existing load/synthesis semantics preserved:
  • POST /model/load may return 202 Accepted while loading.
  • POST /synthesize and /synthesize/stream may return 503 when model is loading.

Developer Experience

• README updated with endpoint docs and curl examples for the new subset.
• Remaining adapter/job endpoints captured in TODO for next iteration.

v0.2.0

Added

• OpenAPI export script: scripts/export_openapi.py.
• Committed OpenAPI spec: openapi/openapi.yaml for easier client integration.
• API docs section in README with live endpoints:
  • /openapi.json
  • /docs
  • /redoc

Improved API Contract

• OpenAPI now explicitly documents:
  • POST /model/load async 202 Accepted loading state.
  • POST /synthesize binary audio/wav response schema.
  • loading/error response semantics for client retry handling.

Why This Release

This release improves developer experience for external clients by making the API contract directly shareable, versionable, and easier to generate typed clients from.

v0.1.1

Fixed

• Fixed launchd runtime PATH so Homebrew-installed sox is found reliably.
• Added startup warm-load support (QWEN_TTS_WARM_LOAD_ON_START) to reduce first-request timeouts.
• Added explicit loading-state behavior:
  • POST /model/load returns 202 Accepted while model loading is in progress.
  • POST /synthesize returns 503 Service Unavailable with Retry-After when model is still loading.

Changed

• GET /model/status now reports loading and load_error fields for better observability.
• Updated launchd/run-service docs and environment examples.

v0.1.0

Added

• FastAPI service scaffold with GET /health.
• Runtime readiness endpoint GET /model/status.
• Lazy model loading endpoint POST /model/load.
• Working synthesis endpoint POST /synthesize returning audio/wav.
• Qwen VoiceDesign smoke script at scripts/voice_design_smoke.py.
• README quickstart, roadmap, and TODO sections.

Changed

• Dependency set updated for Qwen TTS support with qwen-tts and soundfile.
• transformers pinned to 4.57.3 to satisfy qwen-tts requirements.

Notes

• End-to-end synthesis validated via curl and a separate Swift CLI client.