Spaces:
Sleeping
Sleeping
Commit History
-
Enhance ASR and feedback mechanisms with Vietnamese pronunciation tips
- Added VietnameseTipsProvider for detailed pronunciation guidance. - Implemented audio quality assessment and error handling in EnhancedWav2Vec2CharacterASR. - Improved transcription process with retry mechanism and audio validation. - Enhanced feedback generation with specific tips for common phoneme errors. - Updated error handling in ProductionPronunciationAssessor to provide more helpful suggestions. - Refactored existing methods to utilize new Vietnamese tips for better user experience.f784bd2 -
update: new model xlsr
54a64d4
-
feat: Implement ultra-optimizations for pronunciation assessment system
- Removed singleton pattern for thread safety and independent instances. - Enhanced G2P object reuse to avoid redundant creation. - Introduced smart parallel processing based on text length and system resources. - Optimized LRU cache sizes for word and text phoneme lookups. - Added pre-computed phoneme dictionary for common English words. - Implemented object pooling for G2P and comparator instances. - Added batch processing capability for efficient multiple assessments. - Introduced lazy loading for heavy dependencies to improve performance. - Implemented audio feature caching based on file modification time. - Enhanced resource management with intelligent threading strategies. Performance improvements target 80-85% faster processing time, achieving significant reductions in overall assessment duration.225134a -
feat: Implement Whisper model preloading during FastAPI startup for optimized performance
- Added lifespan context manager to preload Whisper model on application startup. - Updated create_app function to include lifespan for preloading. - Enhanced health check endpoint to verify Whisper model loading status. - Refactored speaking_route to export preload function and added documentation. - Optimized post-assessment processing with asynchronous tasks for improved performance. - Created example application demonstrating Whisper preloading integration. - Updated performance test cases to reflect new API endpoint structure.cc06ed6
-
feat: Implement performance optimizations in speaking_route.py
- Added asynchronous processing for post-assessment tasks to reduce processing time. - Introduced shared instances for G2P and ThreadPoolExecutor to improve resource management. - Implemented caching for G2P results to avoid redundant computations. - Enhanced IPA assessment processing with parallel execution for character analysis, phoneme scoring, and focus phonemes analysis. - Created a performance testing script to validate optimizations and measure improvements. - Documented optimization strategies and performance metrics in PERFORMANCE_OPTIMIZATION.md.c9fd875 -
Enhance Vietnamese feedback generation with actionable insights and specific improvement strategies. Refine overall feedback based on score ranges, provide detailed guidance for problematic words and phonemes, and suggest clear next steps for users to improve their pronunciation skills.
b9c5d04 -
Refactor code structure for improved readability and maintainability
85fa45c
-
Implement enhanced pronunciation assessment system with Wav2Vec2 support
- Added Wav2Vec2CharacterASR class for character-level ASR using Wav2Vec2 model. - Integrated SimpleG2P for grapheme-to-phoneme conversion. - Developed WordAnalyzer for analyzing word-level pronunciation accuracy. - Created PhonemeComparator to compare reference and learner phoneme sequences. - Introduced SimpleFeedbackGenerator for generating actionable feedback in Vietnamese. - Implemented SimplePronunciationAssessor as the main interface for pronunciation assessment. - Added test scripts for backward compatibility and enhanced features. - Verified enhanced features and method availability in the new system.aa2c910
-
feat: refactor Wav2Vec2 character ASR to support quantization and improve model loading
c5ca6dc
-
feat: implement Wav2Vec2 character-level ASR with ONNX and Transformers support, add phoneme comparison and feedback generation
cef1d4a -
fix: update import statement for BaseModel in agent.py and add timing logs in speaking_controller.py
1a5420f -
Add Wav2Vec2 model conversion and inference to ONNX format
- Implemented Wav2Vec2ONNXConverter class for converting Wav2Vec2 models to ONNX format, including model loading, conversion, and verification. - Added Wav2Vec2ONNXInference class for performing inference using the converted ONNX model. - Included methods for softmax calculation and transcription of audio files. - Added utility functions for creating compatible models and exporting with fallback options for different ONNX opset versions. - Introduced optimization function for ONNX models. - Created helper function to convert numpy types to native Python types for better compatibility.df380ff