Command-line interface

Audiotext can also be used from the command line, e.g. to transcribe from scripts. It has three commands:

• transcribe: transcribes a file, the files of a folder or a YouTube video.
• watch: transcribes the files added to a folder until it’s stopped with Ctrl+C.
• check-update: checks whether a new version is available and prints the link to download it.

The options that are not given take the values configured in the app. The transcriptions are always saved next to each transcribed file, or in the folder given with --output-dir (where the subfolders of a transcribed folder are recreated).

Run it

The installed app runs the command line when it’s given one of its commands:

System	Command
Linux	audiotext (installed in ~/.local/bin by install.sh)
macOS	/Applications/Audiotext.app/Contents/MacOS/Audiotext
Windows	audiotext-cli.exe, in the folder where Audiotext is installed (%LOCALAPPDATA%\Programs\Audiotext by default)

When running it from the source code, run uv run src/cli.py instead.

Examples

# Transcribe a file. The text is also printed, so it can be redirected
audiotext transcribe interview.mp3 --language es --output-types txt,srt

# Transcribe the files of a folder identifying the speakers
audiotext transcribe recordings/ --diarize --speakers 2 --output-dir transcriptions/

# Transcribe a YouTube video with the Whisper API
audiotext transcribe "https://www.youtube.com/watch?v=…" --method whisper-api

# Transcribe a meeting with the Whisper API, with its keywords and its context
audiotext transcribe meeting.m4a --method whisper-api \
    --openai-model gpt-transcribe --keywords "Audiotext, WhisperX" \
    --prompt "A meeting about the next release"

# Save the subtitles and a Word document, summarize it, and translate it into Spanish
audiotext transcribe talk.mp4 --output-types srt,docx --summarize --translate-to es

# Transcribe the files added to a folder until stopped with Ctrl+C
audiotext watch inbox/ --output-types srt

# The same, transcribing first the files it already contains
audiotext watch inbox/ --existing

Options

Option	Description
-m, --method	Transcription method: whisperx, whisper-api or google
-l, --language	Language of the audio as an ISO 639-1 code (e.g. es), or auto to detect it (not supported by Google)
-t, --output-types	Comma-separated output file types, from aud, json, srt, txt, tsv, vtt, md, docx and ass (e.g. txt,srt,md)
-o, --output-dir	Folder where the transcriptions are saved (default: next to each transcribed file)
--overwrite	Transcribe again the files that already have a transcription, replacing it
--existing	(watch only) First transcribe the files that the folder already contains and that don’t have a transcription yet
-p, --prompt	What the audio is about, such as its topic or setting (not supported by Google)
-k, --keywords	Comma-separated names, terms or acronyms said in the audio, so they’re spelled right (not supported by Google)
--translate	Translate the audio into English while transcribing it (not supported by Google). See --translate-to for other languages
-q, --quiet	Only print errors
-v, --verbose	Print the logs to debug errors
--version	Print the version of Audiotext

The output types that the transcription method doesn’t save itself (e.g. md or docx) are exported from its result, like from the app. aud is only saved by WhisperX.

Summary and translation options

Option	Description
--summarize	Summarize each transcription, with its key points and chapters, into a .summary.md file
--summary-provider	Provider of the language model that summarizes: openai, anthropic, deepseek, gemini, mistral, grok or ollama
--summary-model	Language model that summarizes
--translate-to	Translate each transcription into a language (an ISO 639-1 code), saved in its output file types (e.g. talk.es.srt)
--translation-provider	Provider of the translations: the ones of --summary-provider, deepl or google
--translation-model	Language model that translates, for the providers of language models

The providers and the models are the ones configured in the app unless given. See Providers.

WhisperX options

Option	Description
--diarize	Identify the speakers
--speakers	Number of speakers when identifying them (0 to detect it)
--model-size	Model, e.g. small or large-v2 (see Engines)
--compute-type	int8, float16 or float32
--batch-size	Batch size
--cpu	Run on the CPU, instead of an NVIDIA GPU or the GPU of a Mac

Whisper API options

Option	Description
--openai-model	Transcription model: whisper-1, gpt-transcribe or gpt-4o-transcribe-diarize

Run audiotext transcribe --help to see all the options and their values.

Output and exit code

The progress is printed to the standard error (hide it with --quiet), and the text of the transcription of a single file to the standard output. The command exits with code 1 if a transcription, a summary or a translation fails, and with 130 if it’s stopped with Ctrl+C.

If the API key of the provider of --summarize or --translate-to isn’t set, the command stops before transcribing. A summary or a translation that fails is reported without losing the transcription, which is saved anyway.

The API keys are the ones set in the app, or their environment variables, such as OPENAI_API_KEY, GOOGLE_API_KEY and HF_TOKEN (for identifying the speakers).