Voice transcription for Emacs using OpenAI’s Whisper API.

Requirements

• Emacs 27.1 or later
• ffmpeg
• PulseAudio (currently only supports Linux systems with PulseAudio)
• OpenAI API key

Installation

MELPA

Coming soon.

Manual

Clone this repository and add to your load-path:

(add-to-list 'load-path "/path/to/ok-whisper")
(require 'ok-whisper)

Configuration

Set your OpenAI API key:

(setq ok-whisper--openai-token "your-api-key")

Usage

M-x ok-whisper--record-dwim

Start/stop recording (press once to start, again to stop)

M-x ok-whisper--transcribe-file

Transcribe an existing audio file

The transcription result will be inserted at point.

Define a shortcut:

(define-key global-map (kbd "C-x R") 'ok-whisper--record-dwim)

Limitations

• Currently only works on Linux systems with PulseAudio
• Maximum recording length is 180 seconds by default (but can be configured with ok-whisper--max-recording-seconds
• Requires an active internet connection
• Uses OpenAI’s API (costs apply)

License

AGPL-3.0. See LICENSE.

Credits

Inspired by Simon Safar’s blog post.