ThiThanhChuong/precision-voice

Fixing Notebook Metadata Error

Eval model Whisper

Eval model Whisper

refactor: Adopt diarize-first workflow from notebook

convert the audio to a standard 16kHz Mono WAV format using ffmpeg

Refactor `process_audio` to strictly follow a Diarization -> Segmentation -> Transcription workflow

feat: Add advanced VAD and Whisper generation parameters to the transcription interface.

feat: Pre-load all Whisper models at startup and refactor audio processing to return both transcription and diarization results.

feat: Add dynamic Whisper model loading with caching, enhance diarization with speaker assignment and segment merging, and introduce a dedicated transcription-only function.

feat: integrate Pyannote for speaker diarization, replacing SpeechBrain and adding HF_TOKEN instructions.

feat: Implement automatic speaker diarization and real-time progress updates for transcription and diarization.

feat: remove dedicated Colab notebook and update the simple notebook.

feat: add Vietnamese Speech-to-Text notebook using faster-whisper and Gradio, and update gitignore.

feat: Upgrade speech enhancement to SpeechBrain SepFormer DNS4, integrate Silero VAD, and update transcription model to EraX-WoW-Turbo-V1.1-CT2.

feat: Integrate Silero VAD for pre-transcription silence filtering and timestamp reconstruction.

refactor: Adjust logging verbosity to debug for internal service operations and update orchestrator step descriptions and vocal separation model name.

feat: Implement a denoiser service, switch vocal separation to MDX-Net, and simplify noise reduction.

docs: Add Hugging Face Spaces metadata and configuration reference to README.

ci: Add GitHub Actions workflow to sync repository to Hugging Face Hub.

feat: Introduce Demucs-based vocal separation into the audio processing pipeline with a new service, configuration, and environment variables.

feat: Make application port configurable and add a Google Colab notebook for deployment.

feat: Update Speech-to-Text model to `kiendt/PhoWhisper-large-ct2` in documentation and configuration.

feat: Re-enable and configure VAD, add audio normalization and advanced noise reduction, and implement post-processing for segment merging and filtering.

feat: Update default Whisper model and add a script to verify model configuration.

feat: Implement a real-time processing timer display with associated UI, logic, and styling.

refactor: Transition transcription processing from SSE streaming to a single JSON response, moving progress updates to server-side logs.

feat: Implement real-time processing progress via SSE, add optional noise reduction, and update diarization for pyannote.audio 4.x.

feat: Implement initial Speech-to-Text and Speaker Diarization web application with FastAPI, services, and Docker setup.