Hume Voice Demo

Overview

A Streamlit-based voice agent that lets you hot-swap the underlying language model — Claude or GPT-4 — without restarting the session. Built as a teaching demo to make the differences between models audible, not just measurable.

What you can do

• Speak into the mic; Hume transcribes
• The active model generates a response
• Hume TTS speaks it back
• A circular-spectrum waveform visualizes the audio output in real time
• A single click swaps between Claude and GPT-4 mid-conversation

Why a swap matters

Most voice demos pin you to one model. The point of this one is comparison — same prompt, same TTS, only the brain changes. Useful for showing students or non-technical stakeholders that "the LLM" is not one fixed thing.

Stack

• Python · Streamlit for the UI
• Hume API for transcription + TTS
• OpenAI · Anthropic SDKs with a thin abstraction so swap is a single state change