This is an Ubuntu tool that provides a hotkey to start audio transcription from your microphone, using the whisper_streaming server. The transcribed text will be typed into whatever app you have focused at the current cursor position.

You must first setup and run a whisper_streaming server on either your own machine or another local machine with a good GPU. I've forked the original whisper_streaming repo and made some changes to make it easier to setup and run. To set up the server as a systemd service that will run on system boot, follow my setup instructions.

Then, on your Ubuntu computer, you can use this tool to give you a hotkey that'll stream audio to the whisper server and stream the text into whatever app you have focused at the current cursor position.

Note: This tool has only been tested on Ubuntu 22.04 with the standard GNOME desktop environment, with Python 3.10.12. It may work on other versions of Ubuntu or other Linux distributions, but it has not been tested.

git clone this repo cd into it, then run:

# Feel free to read this script before running it.
# It will apt install some packages and pip install some python packages in a venv in this repo directory.
chmod +x install.sh
./install.sh

You should now see a microphone icon appear in your toolbar. Click on it to open the menu, which provides several options:

• Toggle Recording: Start/stop transcription (same as using the hotkey)
• Settings: Configure server connection, hotkey, and recording duration
• Transcript History: View and copy previous transcriptions
• Restart/Quit Service: Control the whisper-client service

All settings can be configured through the Settings dialog in the menu:

• Whisper server IP address and port number
• Hotkey combination (default: Ctrl+Alt+R)
• Maximum recording duration in seconds (default: 60)

The service automatically starts when you log into Ubuntu. To start audio transcription:

1. Press the hotkey (default: Ctrl+Alt+R) or use the menu's Toggle Recording option
2. Speak into your microphone
3. The transcribed text will appear at your cursor position
4. Press the hotkey again to stop recording, or let it stop automatically after the configured duration

The toolbar icon shows:

• Recording status and duration (e.g., "12/60s" while recording)
• Server connection status
• Time since last server contact

All transcribed text is automatically saved and can be accessed through the Transcript History option in the menu. The history viewer shows:

• Timestamps for each recording session
• Full text of each transcription
• Copy buttons to easily copy any transcript to the clipboard

Transcripts are stored in ~/whisper-transcript.json for future reference, even if they weren't successfully typed into an application.

You can find the service definition file at ~/.config/systemd/user/whisper-client.service if needed. To see the logs run journalctl --user -u whisper-client.service -f. If you make changes to the service file, you can reload it with systemctl --user daemon-reload. To restart the service run systemctl --user restart whisper-client.service.

• Automatically detect when there is no valid microphone connected and disable the microphone icon and hotkey.
  • The icon should change to the "no microphone" icon.
• Add a new dropdown menu item that opens a new dialog, which should show a table:
  • col1 = names of the *.wav files in the recording_path
  • col2 = transcripts of each file (if exist)
  • col3 = button to transcribe the wav file and add the result to the transcript.

The data for this table should be persisted to recording_path/metadata.json We might be able to use the whisper server to do the transcription, but I'm not sure. We might have to use whisper proper, in which case I might wanna just do it on the local machine. TBD.