Enhance speaker assignment in transcription: Introduced interval overlap calculations and smoothing techniques for improved accuracy in speaker labeling. Added methods for determining dominant speakers and stabilizing segment boundaries.

enable result printing and comment out text cleanup regex

model control

remove prompt

update prompt

add prompt

add prompt

fix bug

disable batch

speech duration

less speech duration

enable vad

Refine VAD parameters and transcription options in WhisperTranscriber for improved audio processing. Adjust max speech duration, min speech duration, and silence duration, and set chunk length to 12 seconds.

update log, no vad

print log

add log

disable checksum

update

fix upload issue

upload data

Add job_id and task_id handling in WhisperTranscriber to improve metadata management during audio processing. Update file key generation for intermediate uploads.

fix

add upload

Comment out debug print statements in WhisperTranscriber to clean up output during transcription and embedding calculations.

Enhance waveform processing in WhisperTranscriber by implementing zero-padding for short audio segments and refining segment slicing logic. Update comments for clarity on embedding requirements and fallback mechanisms.

Update waveform handling in WhisperTranscriber to maintain channel dimension during embedding calculations. Adjust comments for clarity on input shape requirements.

Add error handling in embedding calculation within WhisperTranscriber. Log exceptions and diarization segments for better debugging.

try add token

try add token

add log

Implement audio preprocessing and speaker diarization enhancements in WhisperTranscriber. Introduce methods for audio chunk preparation, VAD-based trimming, and speaker embedding extraction. Update process_audio methods to utilize task JSON for improved workflow and metadata handling. Add webrtcvad dependency for voice activity detection.

print result

Add batched inference support in WhisperTranscriber for improved transcription performance. Update methods to accept batch size parameters and adjust output formatting accordingly.

Update transcription batch size to 24 in WhisperTranscriber for improved processing efficiency

Add model downloading functionality for Faster Whisper in app.py, enabling efficient local caching and improved model loading performance.

Refactor model loading in app.py to return both Whisper and diarization models, enhancing GPU utilization during transcription processes.

Refactor WhisperTranscriber to use pre-loaded models instead of loading them during transcription, improving performance and reducing overhead.

Remove model downloading functionality from app.py, reverting to a fallback model name for initialization.

Update Whisper model repository reference in app.py to use OpenAI's whisper-large-v3-turbo

Add model downloading functionality and update GPU initialization process in app.py

init

init without gpu

int8

Update Whisper model compute type to float16 and adjust transcription batch size to 24 for improved performance

Update Whisper model configuration to use int8 compute type for improved performance

print result

Add full audio transcription functionality and update Gradio interface

Enhance error handling for nvidia-cudnn-cu12 integration in app.py. Added checks for the presence of the library and improved loading mechanism with appropriate error messages.

refactor diarizer to remove FP16 model conversion and related configurations

comment out FP16 model conversion in diarizer