managed to make Whisper work with my laptop via a bash script

[rexendevar]

I'm speaking to you now using Whisper on my laptop. Basically, it's using arecord to record the audio of the default sound device into an MP3 file, which Whisper then analyzes. For the most part, integrating whisper itself was fairly easy. The hardest part of it by far was figuring out how to stop arecord.

#!/bin/bash
gunchislas() (
    echo $BASHPID > [scripts path]/pid.txt
    arecord -f cd -D "default" --file-type raw | lame -r - "[scripts path]/.stt.mp3"
)
waiting() {
    zenity --info --text="Taking speech to text input now."
}
gunchislas & waiting & wait -n
pkill -P $(cat [scripts path]/pid.txt) --signal 9
aplay "[scripts path]/clock.wav"
whisper "[scripts path]/.stt.mp3" --language en --model tiny | cut -c 28- | awk NF=NF RS= OFS=" " | xclip
rm "[scripts path]/.stt.mp3"
rm "[scripts path]/pid.txt"
aplay "[scripts path]/recorded.wav" & xdotool type "$(xclip -o)"

I simply press the keyboard shortcut, and it pops up a little window that says "taking speech to text input now," and when I'm done speaking I simply press enter, ensure that my cursor is on the correct text field, and wait. It's awesome.

When I was in middle school I had a speech-to-text program called Dragon or something like that and it was utterly abysmal. I would spend minutes on minutes on minutes fighting with it to say one word and getting so sidetracked and stunlocked. And this is just so much better. This is a thousand miles ahead of whatever dragon was on. I almost just want to waffle on for ages and ages just saying random words because I sincerely love whisper so far. This is insanely good. I only had to correct probably four or five typos so far.