PTE Academic Simulator

A free, browser-based practice simulator for the PTE Academic English exam, covering all 20 official task types with real audio recording and AI-powered scoring. Built with vanilla JavaScript — no framework, no backend, no build step.

🔗 Live demo: https://alansignetti.github.io/pte-simulator/

---

What it is

This tool simulates the four sections of the PTE Academic exam:

• Speaking — Read Aloud, Repeat Sentence, Describe Image, Re-tell Lecture, Answer Short Question
• Writing — Summarize Written Text, Write Essay
• Reading — Multiple Choice, Re-order Paragraphs, Fill in the Blanks (×2)
• Listening — Summarize Spoken Text, Highlight Correct Summary, Write from Dictation, and 5 more

Subjective tasks are scored by Claude using rubrics modelled on Pearson's official 0-90 scoring scale. Objective tasks (MCQ, fill-in-blanks, etc.) are scored locally with deterministic rules.

Quick start

The simulator runs entirely in the browser. You have three options for using it:

Option 1 — Use the hosted version (easiest)

Open https://alansignetti.github.io/pte-simulator/ in Chrome or any Chromium-based browser (Edge, Brave, Arc). On first visit, you'll be prompted to pick a scoring mode:

• Manual scoring (free) — paste prompts into claude.ai using your existing plan (Free, Pro, Team — any works).
• API scoring (paid) — paste your own Anthropic API key, costs ~AUD 0.15 per mini-exam.
• Objective tasks only — skip both; you can still practise MCQ, FIB, Re-order, and Write-from-Dictation tasks for free.

Option 2 — Run it locally

git clone https://github.com/alansignetti/pte-simulator.git
cd pte-simulator
python3 -m http.server 8080
# open http://localhost:8080 in Chrome

Any static web server works (npx serve, Caddy, nginx, etc.). The simulator needs to be served over HTTP — opening index.html directly won't work because ES modules and microphone access require an HTTPS or localhost context.

Option 3 — Deploy your own copy to GitHub Pages

# Fork this repo, then in your fork:
# Settings → Pages → Source: "Deploy from a branch" → main / (root) → Save

After 1–2 minutes your fork will be live at https://<your-username>.github.io/pte-simulator/. No build step needed — GitHub Pages serves the files as-is.

Features

• ✅ All 20 PTE task types with realistic timing
• 🎙 Microphone recording via MediaRecorder + live transcription via Web Speech API
• 🔊 Text-to-speech for Listening tasks using your system voices
• 🤖 AI-powered scoring for subjective tasks using PTE-style rubrics
• 🎲 Unlimited content — "Generate" button on each task creates fresh items with Claude
• 📊 Progress tracking — per-task averages, best scores, sparkline trends, weak-spot detection
• 💸 Zero-cost manual mode — paste prompts into claude.ai instead of using the API
• 🎯 9 mini-exam variants including Classic, Quick Start, Speaking Focus, Drill, Random Mix, and the Full 20-task exam
• ✏️ Editable transcripts — listen back to your recording and fix recognizer errors before scoring
• 🗺 7 chart types for Describe Image (bar, line, pie, table, process flowchart, map of Australia, combo bar+line)

Browser support

Browser	Speaking	Listening	Reading/Writing
Chrome	✅ Full	✅ Full	✅ Full
Edge	✅ Full	✅ Full	✅ Full
Brave/Arc	✅ Full	✅ Full	✅ Full
Safari	⚠️ Partial	✅ Full	✅ Full
Firefox	❌ No SR	✅ Full	✅ Full

Speech recognition is the limiting factor — Firefox lacks the API entirely, Safari's implementation is unreliable. Reading, Writing, and Listening work in every modern browser.

Tech stack

• JavaScript (ES2022, native ES modules)
• HTML5 with <dialog> elements
• CSS3 (no preprocessor)
• Inline SVG for chart rendering
• Anthropic Claude API for scoring and content generation

Browser APIs used

• MediaRecorder (audio capture)
• SpeechSynthesis (TTS for Listening tasks)
• webkitSpeechRecognition (live speech-to-text)
• AudioContext / AnalyserNode (audio level meter)
• localStorage (settings, history, generated content)
• Clipboard API (manual scoring copy)
• HTMLDialogElement (modals)

Architecture

index.html (shell + dialogs)
    │
    ▼
app.js — home, router, exam mode, settings, dashboard
    │
    ├─→ tasks.js     (20 renderers + editable transcript step)
    │       └─→ chart.js (SVG chart renderers)
    ├─→ scoring.js   (PTE rubrics + local scoring)
    │       └─→ api.js  (Claude API or manual paste-dialog)
    ├─→ audio.js     (TTS + MediaRecorder + STT with auto-restart)
    ├─→ history.js   (localStorage-backed score history)
    ├─→ generation.js (per-task AI content generators)
    └─→ content.js   (static question bank + task metadata)

The app uses no global state container — modules communicate through plain function imports.

Project structure

pte-simulator/
├── index.html        # Shell with dialogs
├── styles.css        # All styles
├── app.js            # Home, router, settings, exam, dashboard
├── tasks.js          # The 20 task renderers
├── content.js        # Static question bank + task metadata
├── scoring.js        # PTE-style rubrics + local scoring rules
├── api.js            # Claude API client (with manual-mode fallback)
├── audio.js          # TTS / recording / speech recognition
├── chart.js          # SVG chart renderers (bar/line/pie/table/process/map)
├── history.js        # localStorage score tracking
├── generation.js     # AI-powered content generation per task
├── LICENSE           # MIT
├── .nojekyll         # Disables Jekyll processing on GitHub Pages
└── README.md         # This file

Security & privacy

• The simulator runs entirely in your browser. Nothing is sent to any server except (optionally) directly to Anthropic's API when API mode is enabled.
• Your API key is stored only in localStorage. It never leaves your browser. Anyone who shares your browser profile has access to it — clear it via Settings if you're on a shared machine.
• Score history and generated content are stored in localStorage and stay on your device.
• The Anthropic API call uses the anthropic-dangerous-direct-browser-access header, which is acceptable for this single-user, local-context use case.

Limitations

• Speech recognition isn't perfect — non-native accents may be transcribed with errors. The editable-transcript step lets you fix this before scoring.
• AI scores are approximations of PTE scoring, not official results. They use rubrics modelled on Pearson's published criteria but the actual scoring algorithm is proprietary.
• TTS voices are whatever your operating system provides. For the most realistic experience, use a high-quality system voice (on macOS: Karen, Daniel, Samantha).

Contributing

Contributions are welcome. Things I'd love help with:

• More question content in content.js
• Better TTS integration (ElevenLabs adapter?)
• Spaced-repetition scheduling
• Calibration against real PTE results

Open an issue or pull request on GitHub.

Licence

MIT — see LICENSE.

Author

Alan Signetti — Frontend / Full-Stack Developer on the Sunshine Coast, Australia. Built this while preparing for the PTE myself as part of a skilled-migration pathway. If this saves you some money or time on prep tools, or helps you in your own migration journey, that makes me happy.

• LinkedIn: https://www.linkedin.com/in/alan-signetti-demand
• GitHub: https://github.com/alansignetti
• Portfolio: https://www.alansignetti.com