scrit

A simple cli tool to transcribe audio you are already listening to.

Usage

When %> scrit runs in a terminal:

• It switches audio devices to capture the system output, without interrupting listening.
• It loads OpenAI's Whisper speech-to-text model
• It continuously transcribes the audio and presents a live-updating transcript
• Upon-receiving ctrl-c, it saves the final transcript to file, or copies it to the clipboard

Installation

1. Install BlackHole or another audio loopback program
2. Set up a multi-output devices to add loopback to the audio output devices you want to use

   • detailed here for BlackHole

   • I found it helpful to name each multi-output device "{Audio Device} (tapped)"

   • Here is the side bar of Audio MIDI Setup for my computer

     

3. install scrit with
   • pip install scrit
4. set up the mapping between tapped and un-tapped audio devices to enable automatic switching.
   • edit the scrit configuration with scrit --edit_config
   • in the tap_map section of the configuration file, edit the device names to represent the mapping between normal and tapped versions of the output devices

    "tap_map": [
        {
            "output_device": "External Headphones",
            "tapped_output_device": "External Headphones (tapped)"
        },
        {
            "output_device": "MacBook Pro Speakers",
            "tapped_output_device": "MacBook Pro Speakers (tapped)"
        },
        {
            "output_device": "AirPods",
            "tapped_output_device": "AirPods (tapped)"
        }
    ]

Acknowledgments

• This tool is based on https://github.com/davabase/whisper_real_time
• Transcription is performed by the Whisper speech-to-text model from OpenAI (https://github.com/openai/whisper)
• This tool relies on audio loopback performed by BlackHole but other tools may also work