Update speaker diarization model and refactor WhisperTranscriber alignment process. Introduce align_timestamp method for improved word-level alignment and streamline segment handling. Adjusted print statements for clarity and removed unnecessary comments.

fix params

test word_timestamps=True,

test word_timestamps=True,

fix audio param

fix audio param

log

Remove 'batch_size' from fw_kwargs in WhisperTranscriber to streamline transcription options.

Enhance audio transcription by adding support for 'faster_whisper' engine alongside 'whisperx'. Implement lazy loading for both transcription models and improve handling of transcribe options. Update transcribe_full_audio method to accommodate engine selection and adjust alignment process accordingly.

disable deletion for test

modify out dir

log audio path

segments log

add log

modify preload logic

switch to whisperX

preload

cuda check

remove cuda check

restore to whisper

switch transcribing back to faster_whisper

add log

transcribe

model load

preload

preload

cuda on init

switch to whisperX

fix typo

clean up

modify params

Refactor audio processing: Simplified the handling of audio chunks in prepare_and_save_audio_for_model and updated preprocess_from_task_json to support both single and multiple chunk tasks, enhancing flexibility in audio preparation.

fix field

fix field key

Refactor transcription methods to return results: Updated the transcribe_chunk and transcribe_segments methods to return their results instead of processing them directly, improving the flow of data handling in the WhisperTranscriber class.

update params

add fields

fix typo

add fields

fix bug

fix bug

download all models on startup

fix value

Add audio diarization task to Gradio interface: Introduced a new button and function for audio diarization, allowing users to process audio with speaker separation. Updated existing button labels for clarity.

Refactor model management and transcription process: Introduced a model registry for easier management of Whisper models, added functionality to download models on startup, and streamlined the audio processing pipeline to support both chunk and segment transcriptions with improved error handling and cleanup.

unmatched_diarization_segments

disable clip_timestamps

disable unmatched_diarization_segments

update threshold

try use diarization as clip_timestamp