Use whisper anywhere on your system to enter text. Requires the whisper-asr-webservice, or an OpenAI API key.
I am using skhd for the keyboard shortcuts. Though you can use anything that can execute a program with a keyboard shortcut. Also see this example config for skhd.
It is reccomended to use the main_wrapper script to run the program. I am using file IPC, and therefore having a short startup time reduces input lag significantly (I know, using sockets would just be better. Especially because a server could reduce the python startup time. Maybe I will switch to that eventually). The script only runs python if necessary.
This requires that you put the python path option in the config_local.yaml file (see help). E.g. we would add the line
python_path: /opt/homebrew/Caskroom/miniconda/base/envs/system-wide-whisper/bin/python
To use main_wrapper we also need to have yq (the portable command-line YAML processor) available in the path.
- xdotool
- xclip
usage: main.py [-h] [--start] [--stop] [--toggle-recording] [--toggle-pause]
[--abort] [--clear-notifications] [--no-postprocessing]
[--start-lowercase] [--copy-last] [--list-transcriptions]
[--transcribe-last] [--transcribe-file TRANSCRIBE_FILE]
[--list-recordings] [--only-record] [--clipboard] [--config]
[--voice-announcements]
The default config can be picewise overwritten by a config_local.yaml file
placed in the project directory.
options:
-h, --help show this help message and exit
--start Start the recording.
--stop Stop the recording and transcribe it.
--toggle-recording Start the recording if it is not running, if a
recording is running, stop it and transcribe it.
--toggle-pause Pause/Unpause the recording.
--abort Stop the recording and don't transcribe it
--clear-notifications
Clear all notifications
--no-postprocessing Do not process special commands. E.g. don't translate
'new line' to an actual newline.
--start-lowercase The first character will be lowercase (useful for
inserting text somewhere.)
--copy-last Copy the last transcription to the clipboard.
--list-transcriptions
List all past transcriptions.
--transcribe-last Transcribe the last recording.
--transcribe-file TRANSCRIBE_FILE
Transcribe a file. By default look for the transcribed
files in the project directory. If the argument
contains one or more slashes, it is interpreted as an
path argument relative to the current working
directory. E.g. `-t 2023_06_11-12_53_28.mp3` will look
in the recorded files in the audio directory. `-t
./podcast.mp3` will look for a file 'podcast.mp3' in
the current working directory, and transcribe that.
`-t /home/user/recordings/2023_06_11-12_53_28.mp3`
will look for a file '2023_06_11-12_53_28.mp3' in the
directory '/home/user/memo.mp3' or '~/memo.mp3' will
look for a file 'memo.mp3' in the home directory, and
transcribe that.
--list-recordings List the paths of recorded audio.
--only-record Only record, don't transcribe.
--clipboard Don't paste, only copy to clipboard.
--config Edit the config file.
--voice-announcements
Speak outloud a notification for when recording starts
and ends, and similar events such as pausing.