whisper

whisper transcribes the audio of a media file using whisper.cpp ggml models. It is useful both on its own and for generating subtitles that --edit subtitle can then cut on.

Usage:

auto-editor whisper <file> <model> [options]

<model> is a path to a ggml whisper model. You can download them from huggingface.co/ggerganov/whisper.cpp.

Example:

% auto-editor whisper example.mp4 ggml-medium.en.bin

 And so my fellow Americans, ask not what your country can do for you,
 ask what you can do for your country.

By default the transcript is written to stdout as plain text. Use --output to write to a file and --format to choose the representation:

auto-editor whisper example.mp4 ggml-medium.en.bin --format srt -o out.srt

Options

<file>

The input media file. Only the first audio stream is used; the audio is resampled to 16kHz (whisper's preferred rate) before transcription.

<model>

Path to a ggml whisper model, e.g. ggml-small.en.bin or ggml-large-v2.bin.

-l, --language LANG

Set the spoken language instead of letting whisper auto-detect it. Examples: en, ja. (default auto)

--format FORMAT

Set the output format. Choices are text, srt, and json. (default text)

-o, --output FILE

Where to write the transcript. (defaults to stdout)

-tr, --translate

Translate from the source language to English.

-sw, --split-words

Split the output so that each token is at most one word long.

--queue SECS

The maximum amount of audio (in seconds) queued before processing. Must be between 1 and 86400. (default 30)

--vad-model VAD-MODEL

Set a Voice Activity Detection (VAD) model.

--threads N

Number of CPU threads to use for whisper processing. (default 4)

---

Notes

The whisper filter must be enabled in the ffmpeg auto-editor is using. If it isn't, whisper will report that it Could not find whisper filter.