Fetching metadata from the HF Docker repository...
-
fix app.py to remove redundant generate_kwargs in transcription calls and enable demo sharing during launch
michaeltangz
-
fix app.py to enable demo sharing during launch
michaeltangz
-
refactor app.py to update demo launch logic; ensure proper execution within main guard
michaeltangz
-
fix app.py to correct dtype parameter usage in model initialization and pipeline; remove redundant torch_dtype argument
michaeltangz
-
fix app.py to correct variable name from dtype to torch_dtype in model and pipeline initialization
michaeltangz
-
fix app.py to correct variable name for dtype in model and pipeline initialization
michaeltangz
-
refactor app.py to remove flash attention installation logic and simplify attention implementation; enhance error handling in transcription functions
michaeltangz
-
refactor requirements.txt to remove build isolation for flash attention installation
michaeltangz
-
refactor requirements.txt to ensure flash attention installation without build isolation
michaeltangz
-
refactor requirements.txt to add gradio and numpy dependencies for enhanced functionality
michaeltangz
-
refactor app.py to move flash attention installation into a try-except block; ensure proper fallback to sdpa if installation fails
michaeltangz
-
refactor app.py to streamline flash attention installation and model loading; remove fallback mechanisms and enhance transcription parameters
michaeltangz
-
refactor app.py to remove condition_on_previous_text parameter from transcription functions; enhance output consistency
michaeltangz
-
refactor app.py to enhance pipeline configuration; adjust chunk length for better context and simplify demo interface
michaeltangz
-
refactor app.py to correct parameter name for model loading; streamline microphone input handling
michaeltangz
-
refactor app.py to improve flash attention installation and model loading; add fallback mechanism for model loading
michaeltangz
-
refactor app.py to streamline flash attention installation and model loading; enhance voice activity detection and transcription accuracy
michaeltangz
-
refactor app.py to enhance transcription pipeline by adding condition_on_previous_text parameter; improve transcription accuracy
michaeltangz
-
refactor app.py to specify language parameter in transcription pipeline; enhance accuracy for English transcriptions
michaeltangz
-
refactor app.py to update model loading parameters for flash attention; improve code clarity and maintainability
michaeltangz
-
refactor app.py to remove concurrency limit from audio stream; improve streaming performance
michaeltangz
-
refactor app.py to remove unnecessary time limit from audio stream; enhance streaming functionality
michaeltangz
-
refactor app.py to improve flash-attn installation and model loading with error handling; update demo launch condition
michaeltangz
-
refactor app.py to streamline flash-attn installation and model loading; update requirements.txt to remove unnecessary dependencies
michaeltangz
-
refactor app.py to improve model loading with flash attention and enhance streaming transcription; add packages.txt for dependencies
michaeltangz
-
refactor requirements.txt to remove duplicate entries and ensure proper package versions
michaeltangz
-
add flash-attn to requirements for improved attention mechanisms
michaeltangz
-
install Flash Attention 2 and optimize Whisper model loading; enhance streaming transcription with pipeline approach and latency tracking
michaeltangz
-
update demo launch settings for Spaces to share=False by default
michaeltangz
-
adjust max_new_tokens in transcription functions to leave room for special tokens
michaeltangz
-
enhance audio transcription accuracy with librosa resampling and voice activity detection; increase buffer duration for better context retention
michaeltangz
-
refactor streaming transcription to only emit complete sentences and clear buffer after capture, removing incremental text accumulation logic
-
add incremental text accumulation to prevent duplicate transcription of buffered audio
-
simplify streaming transcription by removing VAD, diarization, and complex buffering logic
-
add speaker diarization with pyannote and optimize streaming pipeline with static caching and deque-based buffering
-
increase buffer duration and adjust transcription parameters to prioritize accuracy over latency
-
reduce buffer duration and adjust transcription parameters for lower latency
-
improve sentence splitting to avoid breaking on common abbreviations like Mr., Ms., Dr., etc.
-
remove live subtitle display and only show full transcript in the UI
-
improve near-duplicate detection to compare against all existing sentences instead of only the last one
-
add near-duplicate detection to suppress streaming variants of the same sentence
-
update live subtitle to persist last sentence when no incomplete audio
-
Merge branch 'asr-1'
-
refactor to use per-session state instead of global variables
-
update the instruction(copy) for ui
-
update the instruction(copy) for ui
-
edit the instruction for ui
-
uodate the new words modification
-
optimize the words for better display on ui
-
change to small whisper model
