Build a Simple Voice Assistant

The pipeline

A voice assistant is three components in a loop:

1. Speech-to-text (Whisper) — transcribe the user.
2. LLM — generate a response.
3. Text-to-speech (ElevenLabs / OpenAI TTS) — speak it back.

text = stt(audio)
reply = llm(text)
speak(tts(reply))

Polish

Stream each stage to cut perceived latency, add a wake word or push-to-talk, and keep conversation history for context. The same pattern scales from a desktop toy to a phone agent.