whisper/transcribe.py: decoding defaults and command-line options (openai/whisper)

“If the no_speech probability is higher than this value AND the average log probability over sampled tokens is below `logprob_threshold`, consider the segment as silent” (Docstring for the no_speech_threshold argument of the transcribe function, in whisper/transcribe.py)

“parser.add_argument("--device", default="cuda" if torch.cuda.is_available() else "cpu", help="device to use for PyTorch inference")” (Command-line options (--device))

“disabling may make the text inconsistent across windows, but the model becomes less prone to getting stuck in a failure loop” (transcribe() parameters: condition_on_previous_text)

“FP16 is not supported on CPU; using FP32 instead” (Model decoding setup)

“a context for transcription, e.g. custom vocabularies or proper nouns to make it more likely to predict those word correctly.” (transcribe() parameters: initial_prompt)

“language spoken in the audio, specify None to perform language detection” (Command-line options (--language))

“the path to save model files; uses ~/.cache/whisper by default” (Command-line options (--model_dir))

“which will be successively used upon failures according to either `compression_ratio_threshold` or `logprob_threshold`.” (transcribe() parameters: temperature)

“(requires --word_timestamps True) the maximum number of lines in a segment” (Command-line argument parser, --max_line_count option)

“(requires --word_timestamps True) the maximum number of characters in a line before breaking the line” (Command-line argument parser, --max_line_width option)

“format of the output file; if not specified, all available formats will be produced” (Command-line options, --output_format)

“choices=["txt", "vtt", "srt", "tsv", "json", "jsonl", "all"]” (Command-line argument parser, --output_format option)

“temperature = tuple(np.arange(temperature, 1.0 + 1e-6, increment))” (Command-line options, after argument parsing)

“(experimental) extract word-level timestamps and refine the results based on them” (Command-line argument parser, --word_timestamps option)

“Performing inference on CPU when CUDA is available” (transcribe(), device check after the model is loaded)

“Detecting language using up to the first 30 seconds” (transcribe() function, language detection block)

“"--no_speech_threshold", type=optional_float, default=0.6, help="if the probability of the” (Command-line options, argparse definition of --no_speech_threshold)

“If the no_speech probability is higher than this value AND the average log probability over sampled tokens is below ... consider the segment as silent” (transcribe() parameter documentation in whisper/transcribe.py)