Empathic Voice Interface | Chat History

Fetch Chat Events, Generate a Transcript, and Identify Top Emotions

Overview

This project demonstrates how to:

Retrieve all chat events for a specified Chat ID from Hume's Empathic Voice Interface (EVI) using the TypeScript SDK.
Parse user and assistant messages to produce a formatted chat transcript.
Compute the top three average emotion scores from user messages, leveraging the built-in EmotionScores interface.

Key Features:

Transcript generation: Outputs a human-readable .txt file capturing the conversation between user and assistant.
Top 3 emotions: Identifies the three emotions with the highest average scores across all user messages, returning them as a Partial<EmotionScores> object.

Ensure your environment meets the following requirements:

Check versions on macOS:

node -v
npm -v

If you need to update or install Node.js, visit the official Node.js website.

Obtain Your API Key: Follow the instructions in the Hume documentation to acquire your API key.
Create a .env File: In the project's root directory, create a .env file if it doesn't exist. Add your API key:

HUME_API_KEY="<YOUR_API_KEY>"

Refer to .env.example as a template.

In the main function within src/index.ts, set the CHAT_ID variable to the target conversation ID:

async function main(): Promise<void> {
  const CHAT_ID = "<YOUR_CHAT_ID>"; // Replace with your actual Chat ID
  // ...
}

This determines which Chat's events to fetch and process.

npm install

npm run dev

The script fetches all events for the specified CHAT_ID.
It generates a transcript_<CHAT_ID>.txt file containing the user and assistant messages with timestamps.
It logs the top 3 average emotions to the console:

{
  "Joy": 0.7419108072916666,
  "Interest": 0.63111979166666666,
  "Amusement": 0.63061116536458334
}

(These keys and scores are just examples; the actual output depends on the Chat's content.)