add deepspeed

update: new model xlsr

feat: Implement Whisper model preloading during FastAPI startup for optimized performance

feat: Implement performance optimizations in speaking_route.py

feat: implement new IPA assessment API with detailed phoneme analysis and Vietnamese-specific feedback

Refactor code structure for improved readability and maintainability

Implement enhanced pronunciation assessment system with Wav2Vec2 support

Implement Lesson Practice 2 Agent with state management and routing logic; add practice and teaching agent functions; create prompts for conversation and teaching; establish pronunciation API with lesson data and search functionality.

feat: refactor Wav2Vec2 character ASR to support quantization and improve model loading

Add Wav2Vec2 model conversion and inference to ONNX format

Implement code changes to enhance functionality and improve performance

Implement feature X to enhance user experience and fix bug Y in module Z

feat: add speaking route for pronunciation assessment API