Chunk Whisper

This is a simple whisper CLI application that instead of streaming whisper, it will only run the model after a 3 second pause.

Installation

pipenv install

Usage

pipenv run python record.py

It will first prompt you to select a microphone, and after that it will start recording. Pressing enter will stop the recording and run the model on the recorded audio.

If your audio has 3 seconds of silence, it will run the model on the audio and count that as a paragraph.

Configuration

Argument	Description	Default
--model	The whisper model to use. list	`small`
--microphone	Microphone name. The first match that contains this string will be picked. Special options: `pick` will prompt you to pick one. `default` will figure out the first valid one.	`pick` (`default` if --silent is used)
--silent	Don't write anything in the console except for the translated audio, and a "Start" to know when to start talking.	`False`
--pause-seconds	Seconds of silence to wait for making a new chunk (paragraph).	`3`
--debug	Print debug information.	`False`

Example usages

Useful to simply get started with the model.

pipenv run python record.py --silent

To try different models. Check them all here.

pipenv run python record.py --silent --model=base

To use a fixed microphone. Note that --microphone=pick does not work with --silent.

pipenv run python record.py --silent --microphone="MacBook Pro Microphone"

To add the content to a file after running the model. (you won't see the "start" prompt)

pipenv run python record.py --microphone="default" >> file.txt

To-do