Add Qwen3-ASR support with OpenAI-compatible transcriptions endpoint

[juntao]

Summary

• Add Qwen3-ASR model support with /v1/audio/transcriptions endpoint following OpenAI API format
• Rename environment variables for clarity: BASE_MODEL_PATH → TTS_BASE_MODEL_PATH, CUSTOMVOICE_MODEL_PATH → TTS_CUSTOMVOICE_MODEL_PATH
• Rename Docker image from qwen-tts-api to qwen3-audio-api
• Add ffmpeg dependency for audio format conversion
• Fix nagisa SyntaxWarning and prevent ruff download at container startup
• Add Phase 4 (ASR-only) and Phase 5 (TTS→ASR round-trip) tests to TEST_PLAN.md and CI

Test plan

• Build Docker image with docker build -t qwen3-audio-api .
• Run container with TTS and ASR models mounted
• Test TTS endpoint: curl -X POST http://localhost:8000/v1/audio/speech -H "Content-Type: application/json" -d '{"input":"Hello world","voice":"Vivian"}' --output test.wav
• Test ASR endpoint: curl -X POST http://localhost:8000/v1/audio/transcriptions -F file=@test.wav
• Verify round-trip: TTS output text should approximately match ASR transcription
• Verify no nagisa SyntaxWarning on startup
• Verify no ruff download on container start

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR adds Qwen3-ASR (Automatic Speech Recognition) support to the existing Qwen3-TTS API server, transforming it into a comprehensive audio processing API that provides both text-to-speech and speech-to-text capabilities through OpenAI-compatible endpoints.

Changes:

• Implements /v1/audio/transcriptions endpoint following OpenAI API format for speech-to-text functionality
• Renames environment variables for clarity (BASE_MODEL_PATH → TTS_BASE_MODEL_PATH, CUSTOMVOICE_MODEL_PATH → TTS_CUSTOMVOICE_MODEL_PATH) and adds ASR_MODEL_PATH
• Adds new dependencies (qwen-asr, python-multipart, av, nagisa, soynlp) and includes ffmpeg for audio format conversion
• Updates Docker configuration to suppress nagisa SyntaxWarning and prevent ruff download at container startup
• Expands CI test coverage with Phase 4 (ASR-only) and Phase 5 (TTS→ASR round-trip) tests

Reviewed changes

Copilot reviewed 8 out of 9 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
python/uv.lock	Adds dependencies for ASR support (qwen-asr, av, nagisa, etc.) and pins transformers to 4.57.6
python/pyproject.toml	Updates project description and adds qwen-asr, python-multipart dependencies with transformers override
python/main.py	Implements ASR endpoint with audio conversion helpers and model loading logic
python/TEST_PLAN.md	Adds Phase 4 (ASR-only) and Phase 5 (TTS→ASR round-trip) test scenarios
python/README.md	Documents new ASR features, endpoints, and usage examples
python/Dockerfile.cuda	Adds PYTHONWARNINGS environment variable and changes CMD to use venv python directly
python/Dockerfile	Same Docker improvements as CUDA version
README.md	Updates project description to include ASR capabilities
.github/workflows/ci.yml	Adds ffmpeg installation, ASR model downloads, and new test phases

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The parameters 'prompt' and 'temperature' are accepted in the create_transcription function but are not used in the implementation. While this is mentioned in the inline comment as "not currently used", accepting parameters without using them can be confusing for API consumers. Consider either implementing support for these parameters or documenting in the API reference that they are accepted for OpenAI compatibility but currently ignored by the Qwen3-ASR model.