Fix ZeroGPU CUDA init error: force CPU loading, GPU only during inference

Hybrid mode: GPU with CPU fallback when quota exceeded

Update whisper medium scoring_engine.py

Update articulation levels and scoring weights in constants.py

Fix double-run: separate model loading (CPU) from inference (GPU)

Fix ValueError: remove invalid condition_on_previous_text, use temperature 0.2 + num_beams 1

Fix 3 issues: disable Whisper auto-correct, reduce timeout 120s, preload model to prevent double-run

Fix ASR logprobs error - load audio as array, use ISO language code, add logprob_threshold

MAJOR UPGRADE: Whisper Large V3, 30+ Indonesian phonetics, Gradio JSON API, optimized weights

Fix mount_gradio_app: import gradio and correct method call

Add Wav2Vec2 Indonesian for Level 1-3, mount FastAPI to Gradio for Swagger access

Enable Swagger UI/ReDoc and fix pipeline language parameter error

Major fix: Force Indonesian language (id), add phonetic similarity for T/D confusion, improve short audio handling, add WER/Articulation explanations

Fix ZeroGPU duration: 240s -> 180s (max allowed)

Upgrade to Whisper Medium for better accuracy, increase GPU timeout to 240s, add chunk processing for long audio, add quota warning

Fix deprecation warning for Whisper and remove Examples causing IndexError

Major improvements: Fix UI text visibility, improve similarity calculation (remove punctuation), optimize Level 1-3 scoring (remove pitch/speech rate), enhance articulation for vowels

Fix torchcodec error: Replace torchaudio.load with librosa.load for better stability

Fix IndentationError: remove duplicate level=level line

Fix dependency conflict: python-multipart 0.0.6 -> >=0.0.9 for Gradio 5.0.0 compatibility

Critical fix: Add error suppression, pin dependencies, reduce logging to fix Content-Length error

Fix: Disable OpenAPI docs and simplify endpoints to fix Content-Length error

Fix: Use orjson for robust JSON serialization with proper UTF-8 encoding

Add comprehensive Swagger/OpenAPI documentation with detailed endpoint descriptions

Fix: Add JSON sanitization for numpy types to prevent Content-Length errors

Fix: Resolve Content-Length error with explicit JSONResponse serialization

Fix: Update imports and upgrade Gradio to 5.0.0

Deploy Vocal Articulation Assessment v2.0

initial commit