Skip to content

Home

Jump to bottom
Tim edited this page Dec 16, 2025 · 3 revisions

What is Talk2Care

This repository contains the backend for the Comenius Project: Talk2Care. Talk2Care is a conversational AI developed to engage with students and simulate realistic scenarios that they can rehearse before applying them in real-life situations.

Background

Talk2Care was created to explore how effectively AI can be used as an educational tool within study programs. It has been specifically tuned and optimized for the nursing program, where it will be used throughout the testing period. Depending on the outcomes, Talk2Care may also be deployed in other programs to help students practice and improve their conversational skills.

What is this Wiki for

This wiki and repository provide the backend resources for the project. All functionality and relevant information can be found here.

General Overview

The backend consists of 4 key component: A manager, a speech-to-text (STT), a learning machine model (LLM), and a text-to-speech (TTS). These components work together in order to simulate a virtual character that a user can interact with.

  • Speech-to-Text (STT): Uses a model from faster-whisper to convert user audio into text.
  • Large Language Model (LLM): Utilizes the Ollama framework to download and run the latest supported LLM used for generating responses.
  • Text-to-Speech (TTS): Uses Coqui-TTS to convert generated text back into speech.
  • Manager: The only component that does not use a machine-learning model. It acts as the entry point of the backend and orchestrates communication between STT, LLM, and TTS.
|- App
|  |- faster_whisper
|  |- manager
|  |- ollama
|  |- tts
|- README.md
|- client.py
|- docker-compose.yml

There is a fifth optional component, client.py, which serves as a placeholder for the frontend. When connected to an external application, client.py is not required. It exists purely for testing backend functionality and features.

Clone this wiki locally