Skip to main content

npx remotion transcribe

Transcribe an audio or video file to captions using Whisper.cpp. The command handles installation, model downloading, audio conversion, and transcription automatically.

npx remotion transcribe <input-file> [output-file]

The output file defaults to <input-basename>.json if not specified.

The output is a JSON file containing a Caption[] array from @remotion/captions.

Phases​

The command runs through 4 phases, each showing progress:

  1. Install whisper.cpp - Clones and builds whisper.cpp if not already present
  2. Download model - Downloads the specified Whisper model if not already present
  3. Convert audio - Converts the input file to a 16kHz WAV file using FFmpeg
  4. Transcribe - Runs Whisper.cpp on the audio and outputs captions

Example​

npx remotion transcribe video.mp4

This will create video.json with the transcription.

npx remotion transcribe video.mp4 captions.json --model=large-v3

This will transcribe using the large-v3 model and save to captions.json.

Flags​

--model​

The Whisper model to use. Default: medium.

Available models: tiny, tiny.en, base, base.en, small, small.en, medium, medium.en, large-v1, large-v2, large-v3, large-v3-turbo.

--whisper-version​

The version of whisper.cpp to install. Default: 1.5.5.

--whisper-path​

The path where whisper.cpp should be installed. Default: ./whisper.cpp.

--model-folder​

Custom folder for storing Whisper models. Defaults to the models folder inside the whisper.cpp directory.

--language​

Specify the language of the audio. If not set, Whisper will auto-detect the language.

--translate-to-english​

Translate the transcription to English. Default: false.

--token-level-timestamps​

Enable token-level timestamps using DTW. Default: true.

--flash-attention​

Enable flash attention for faster transcription. Default: false.

--log​

Set the log level. Use --log=verbose to see all compilation and transcription output line-by-line instead of progress bars.

--quiet / -q​

Suppress all output.

See also​