FFmpeg 8.0 adds Whisper support
11 days ago
- #whisper
- #audio-filter
- #transcription
- A new audio filter named 'whisper' is introduced for running audio transcriptions using the Whisper model.
- The filter requires the whisper.cpp library and can be enabled with './configure --enable-whisper'.
- Key options include model path, language (default 'auto'), queue size (default 3s), GPU usage (default true), and output format (text, srt, json).
- VAD (Voice Activity Detection) model support is included for better transcription quality with parameters like threshold, min speech/silence durations.
- Examples provided for transcription with SRT file generation, JSON output to HTTP service, and microphone input with VAD.