openai/whisper ? reverse-engineered prompt

I want to build a tool that uses the OpenAI Whisper model for speech-to-text. The main function should be taking an audio file, like an MP3 or WAV, and transcribing what's said into text. It needs to be versatile, so please make sure it supports multiple languages. Also include a feature to translate audio from another language directly into English.

I've heard there are different model sizes available, from 'tiny' and 'base' up to 'large', which offer a trade-off between speed and accuracy. Please set it up so I can easily choose which model to use when I run a transcription.

I'd like this to be a Python project, accessible both as a simple command-line tool and as a module I can import and use in other scripts. Just make sure all the dependencies, like PyTorch and anything else it needs to run, are handled.