This is an implementation of OpenAI's Whisper for the purpose of speech-to-text via your default microphone, enabling direct output to your clipboard and/or CLI. In addition, the script records and inferences with the press of a desired keystroke combination.
Installation is as easy as:
pip install -U openai-whisper
pip install -r requirements.txt
I also strongly encourage the installation of PyTorch with CUDA:
pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu117
If you choose to use PyTorch with your CPU instead, please run:
pip install torch
Most input arguments carry over from the base Whisper package.
Those of note are:
--model default is the 'small' model, some others are the 'tiny', 'base', 'medium', and 'large' models
--model_dir default is set as None, 'Users/[username]/.cache/whisper' is used as the default dir
--device default is 'cuda', set this to 'cpu' if you don't have cuda installed with torch
--task default is transcribe (duh), but translation is also possible
--language default is set as None (enabling language detection), but I recommend setting this to English/French/Spanish/etc. to cut down on inference times
--temperature default is 0, depolarizes output distribution allowing for more 'creativity'
--threads default is 0, this refers to CPU threads
Novel input arguments:
--amplification default is 1.0, amplifies the recording by a given floating point multiple
--print_text default is not set/False, prints text to your CLI/terminal
--no_clipboard default is not set/False, prevents output from being sent to your clipboard
--hotkey default is 'ctrl+shift+r', the combination that activates audio recording and inferencing