Test OpenAPI Whisper: Record and Transcribe via Terminal

output-example.log

[00:00.000 --> 00:17.320]  I'm currently testing Whisper on my desktop as a way to eventually share with the linking
[00:17.320 --> 00:20.080]  your thinking community.
[00:20.080 --> 00:28.720]  This is just a way to show different timestamps and general display ability and seeing how
[00:28.720 --> 00:34.000]  this could look as an easy note taking solution.
[00:34.000 --> 00:38.240]  I for one think this is a rather cool piece of software that I could simply run from my
[00:38.240 --> 00:40.960]  terminal of my Linux desktop.

setup-whisper.sh

# First have pyenv setup
# - https://github.com/pyenv/pyenv

# Create dir for whisper work
mkdir -p ~/git/whisper
cd ~/git/whisper
# Use Python <3.11 because whisper isn't supported in
# Python 3.11.x just yet
echo '3.10.10' > .python-version
pyenv install 3.10.10
pyenv virtualenv 3.10.10 whisper
pyenv activate whisper

# Upgrade pip, install whisper
pip install -U pip wheel
pip install openai-whisper

transcribe.sh

# Expects to be in ~/git/whisper
# Expects setup-whisper.sh has already been used
# Expects ffmpeg to be installed
# cd ~/git/whisper
# pyenv activate whisper

# This line just turns on the microphone to record; then press Ctrl+C when done recording.
# It outputs a recording file called "audio.mp3"
# To find available sound devices, I first ran this command on Linux to list pulse audio devices:
# -> ffmpeg -sources pulse
# I then selected the microphone I wanted to use in '<name-of-my-microphone-device>'
ffmpeg -f pulse -i '<name-of-my-microphone-device>' audio.mp3

# Transcribe the audio into text
# I go with the small model because I have 4GB VRAM
# See readme: https://github.com/openai/whisper
whisper audio.mp3 --model small --language English | tee transcription.log