• The ffmpeg module was replaced with the subprocess module due to conflicting argument name "async"
• A new function get_audio() is added which extracts audio from a video file and returns its path
• In main(), if --extract-audio flag is set, then extract audio using get_audio() instead of calling transcribe().

@RapDoodle Can you update this to use the newer aresample option instead? I believe this will also let us keep using ffmpeg directly

• The ffmpeg module was replaced with the subprocess module due to conflicting argument name "async"
• A new function get_audio() is added which extracts audio from a video file and returns its path
• In main(), if --extract-audio flag is set, then extract audio using get_audio() instead of calling transcribe().

[what-the-diff] what a lovely but completely useless chap you are 😉. Your contribution made no sense to this thread. Anyway, @RapDoodle I'm afraid the de-synchronization still happens despite the changes!

Following the discussion here openai/whisper#1888 I was able to greatly improve accuracy by adding the --word_timestamps True flag to the transcribe call.

GitHub

About timing sync of Ger video translation · openai whisper · Discussion #1888

Hail I'm trying to translate this German movie into English or at least transcribing. I see that the translation or transcribing are truly impressive, nothing to complain. And somehow somewhat, the...