deepgram/deepgram-python-sdk ? reverse-engineered prompt

Build me an official Python SDK for Deepgram that makes it easy to use speech to text, text to speech, text analysis, and voice agent features from one clean client.

I want it to work with both normal and async code, support API key and access token auth from either parameters or environment variables, and feel simple enough that someone can transcribe an audio file, stream live microphone audio, generate speech from text, or analyze a sentence with just a few lines.

Please include good examples, clear error handling, and support for real time WebSocket connections as well as regular request and response calls. Make the package easy to install and use, with solid documentation and tests, and organize it so future API changes are manageable. If you need to check the latest Deepgram docs or API details online, go ahead and do that.