whisperpluck

Some scripts for convenient voice transcription. At present, we use OpenAI-Whisper, since it's such high quality.

Features

• Transcribes your voice, and copies it to the clipboard.

• Can run whisper each time you record yourself

• Can run our little whisper server (in server-flask/ dir) for faster use (model stays loaded)

• UI buttons: In ui/whisper-buttons may be run for on-screen buttons

• Hotkeys: You can assign hotkeys to whisper-auto and whisper-kill-rec

• Desktop Shortcuts: Drag the scripts to your desktop (preferably as .desktop entries -- I'm not sure what they'll do as links because whisper-auto tries to find its own location)

Video of optional UI buttons

Click to play video (remember, the button UI is optional. Hotkeys can be assigned to the scripts, or they can be used as icons on the desktop):

Please consider contributing...

• Donate to patreon.com/jaggz
• Or paypal me at jaggz.h {at} gmail.com

IMPORTANT, Maybe...

This was initially designed for the desktop icons method of execution. It updates/renames icons on the desktop to provide status changes (like, it'll rename them to WH-Ready, WH-Rec, ...). This doesn't work on all X11 desktops (the renaming doesn't update the desktop entries). I worked on this for some time but gave up and focused on the little GTK ui (ui/whisper-buttons).

WHISPER USAGE METHODS

We have two methods of using whisper for our purposes:

1. No-Server Method: Runs whisper each time you want a transcription (model must be loaded each run)
2. Server Method: Run a separate little server (in server-flask/) which keeps the model loaded so transcriptions are faster

Both methods require a venv, which right now is hard-coded as whisper

(A venv is a python virtual environment, to store all the whisper-related stuff in this case.)

1. Make venv:
   1. Make venv base dir: mkdir -p ~/venv (or wherever you want it. Adjust the following commands to match this base directory.)
   2. Create whisper venv: python -m venv ~/venv/whisper
2. Activate venv: . ~/venv/whisper/bin/activate
3. Install whisper stuff in venv:

   1. pip install openai-whisper

   2. OR, you can use the versions I'm currently using, which I placed in requirements-whisper.txt (to just use the scripts or UI), and server-flask/requirements-server.txt (to run the server, which needs openai-whisper AND flask).

   3. pip install flask <--- used only if you want to run our server

Usage without server (will work right away.. least hassle to get started)

1. Edit whisper-auto
2. Change "use_server=1to=0`
3. If you used a venv other than ~/venv/whisper:
   1. You'll need to change the envact= line to point to it
   2. and the envpattern= line which is used to detect if we're in the venv already.
4. Run ./whisper-auto (it will begin recording)
   1. Hit ctrl-c to stop recording, and whisper should begin processing.
   2. Alternatively, use the whisper-kill-rec script to end the recording task.

Usage with our Whisper Server

In this mode, we run a separate little python app that keeps the model loaded.

https://blog.deepgram.com/how-to-build-an-openai-whisper-api/ Implemented an ultra-easy flask server (modified in this here project) for the task. Thanks Adam! Check out his blog post.

1. In a separate term, go to the server-flask folder
   1. flask run should run it fine
   2. It uses port 5000 by default. Edit the app.py and whisper-auto to set the server\_port= if you change it.
2. Run ./whisper-auto (it will begin recording)
   1. Hit ctrl-c to stop recording, and whisper should begin processing.
   2. Alternatively, use the whisper-kill-rec script to end the recording task.

Installation - UI

1. Go to the ui/ folder and type `make`
1. I've not built out a list of dependencies -- you'll have to figure it out yourself (it uses gtk).  Submit a pull-request with dependencies added here if you want :)

Installation - Hotkeys

1. Assign two hotkeys, one to -auto and one to -kill-rec

Installation - Desktop icons

1. Drag both to your desktop, or otherwise add them.

Note: They should probably be links or .desktop entries, not copies, so updates will be handled when you git pull updates.