ProjectBuild an AI Voice Receptionist: the explanation · Module 3: Choosing models and infrastructure · part 5 of 23
Part 5 · Module 3: Choosing models and infrastructure
Module 3: Choosing models and infrastructure
23 min read·
About 40 minutes · Needs: Module 1 (the requirements), Module 2 (the voice loop) · Code: research/stt.py, research/tts.py, backend/app/core/config.py, docker-compose.yml, .env.example
Where this fits
The parts whose technology this module chooses (blue): speech to text, the models, the voice, search and the two data stores.
Module 2 named the stages of a turn. This module chooses what runs each one: which speech recogniser, which language models, which voice, which embedding model, which databases, and where each of them runs. The highlighted parts are the ones a choice here decides. Every choice was made by measuring, in a short experiment before the build and in evaluations after it, and every one can be changed with a setting.
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.