Taylor

AI Voice Coaching Agent (Project by Team Calli)

Real-time English conversation coach designed for Business & Travel scenarios. Powered by OpenAI Realtime API (WebSocket/Async).

Overview

Taylor is a voice-first AI agent developed by Calli, aimed at helping non-native speakers practice English in realistic scenarios. Unlike traditional TTS/STT pipelines, Taylor utilizes OpenAI's Realtime API to achieve ultra-low latency and natural turn-taking.

This project was developed as a Rapid Prototype to solve a personal pain point: the lack of accessible, real-time speaking practice partners.

Key Features

• True Real-time Interaction: Leverages OpenAI Realtime API for fluid, interruption-friendly conversations without the lag of traditional transcoding.
• Scenario-Based Coaching: Supports 15+ roleplay scenarios including Business Negotiation, Hotel Check-in, and Casual Small Talk.
• Direct Audio Streaming: Implemented raw audio streaming logic using Python AsyncIO for seamless bi-directional communication.

Tech Stack

• Core Engine: OpenAI Realtime API (GPT-4o Audio)
• Protocol: WebSockets
• Language: Python (AsyncIO)

How It Works

1. The user starts the conversation with a specific topic (e.g., "Salary Negotiation").
2. Taylor initiates the dialogue with a scenario-specific opening.
3. The system streams audio directly to/from OpenAI, allowing for instant feedback.
4. Taylor responds with natural dialogue, adopting a strict but helpful persona.

Future Improvements

• Dashboard: Visualize "Grammar Accuracy" and vocabulary usage over time.
• Dual Modes:
  • Practice Mode: Strict persona for intensive training.
  • Realtime Mode: Instant assistance for real-world conversations (e.g., translating/hinting during a meeting).

---

Created by Hyuntae Jeong.