ProjectBuild an AI Voice Receptionist: the explanation · Module 9: Speech to text with Whisper · part 11 of 23
Part 11 · Module 9: Speech to text with Whisper
Module 9: Speech to text with Whisper
20 min read·
About 35 minutes · Needs: Module 8 (how the VAD cuts a turn), Module 3 (choosing models and infrastructure) · Code: backend/app/voice/stt.py, backend/app/voice/session.py, backend/app/voice/audio.py
Where this fits
Whisper (blue: this module) takes each finished turn from the VAD and turns it into text and a language for the agent.
The VAD hands over one finished turn of caller audio at a time. Before the agent can do anything, those sounds have to become words. This module sends each turn to Whisper large-v3 on Groq and gets back the text and the language the caller spoke. Everything after it, from language detection to booking, works on that text, so a misheard time here becomes a wrong appointment later.
The rest of this project is yours to keep
This project is bought on its own, once, and stays readable afterwards, including the parts added to it later.