🧪 Model Test

Model Test is the complete research and testing environment used during the development of the Aniquen Wake Word AI.

This repository contains every stage of the wake-word pipeline including browser inference, server inference, microphone recording, offline validation, ONNX model execution, audio preprocessing and debugging utilities.

It is designed for experimenting with keyword spotting models before deployment into production applications.

---

📑 Table of Contents

• Features
• Screenshots
• Technologies
• Project Structure
• Architecture
• How It Works
• Testing Workflow
• Installation
• Future Improvements
• Contributing
• License
• Author

---

✨ Features

✅ Browser Wake Word Detection

✅ Server Wake Word Detection

✅ ONNX Runtime Integration

✅ Mel Spectrogram Generation

✅ Embedding Extraction

✅ Keyword Spotting Model

✅ Live Microphone Streaming

✅ WebSocket Communication

✅ Offline Pipeline Testing

✅ WAV Recording Utility

✅ Automatic Audio Resampling

✅ Float32 PCM Processing

✅ Real-Time Detection Visualization

✅ Debug Probability Logging

✅ Research & Benchmark Environment

---

🏗 Architecture

Microphone
      │
      ▼
Browser
      │
      ├────────────► Recorder
      │
      ├────────────► Offline Test
      │
      ├────────────► Browser Pipeline
      │
      └────────────► WebSocket
                          │
                          ▼
                  FastAPI Server
                          │
                Mel Spectrogram
                          │
                    Embedding Model
                          │
                  Wake Word Model
                          │
                    Detection Result

---

📸 Screenshots

Home Screen

---

Wake Word Detection

---

Offline Pipeline

---

Probability Logs

---

⚙ Technologies Used

• HTML5
• CSS3
• JavaScript (ES6)
• Python
• FastAPI
• WebSocket API
• ONNX Runtime Web
• ONNX Runtime
• NumPy

AI Models

• Mel Spectrogram Model
• Embedding Model
• Hey Aniquen Wake Word Model

---

📂 Project Structure

Model-Test/
│
├── index.html
├── record.html
├── test_offline.html
├── script_server.js
├── style.css
├── recording.wav
│
├── embedding_model.onnx
├── melspectrogram.onnx
├── hey_Aniquen.onnx
│
└── README.md

---

⚡ How It Works

Browser Pipeline

1. Access microphone

↓

2. Capture Float32 PCM audio

↓

3. Resample to 16 kHz

↓

4. Stream audio over WebSocket

↓

5. Receive wake probability

↓

6. Trigger wake-word animation

---

Offline Pipeline

1. Load recording.wav

↓

2. Generate Mel Spectrogram

↓

3. Create Embeddings

↓

4. Run Wake Word Model

↓

5. Print probabilities

↓

6. Verify detection threshold

---

Recording Utility

The recorder captures:

• 16 kHz Mono Audio
• Float32 PCM
• WAV Export
• Six-second Test Clips

These recordings are used for validating the AI models.

---

🧠 AI Pipeline

Audio

↓

Mel Spectrogram

↓

Embedding Network

↓

Context Window

↓

Wake Word Model

↓

Probability Score

↓

Detection

---

🧪 Testing Workflow

The repository supports multiple testing modes.

Live Browser Test

Real-time microphone detection using WebSockets.

---

Offline Validation

Replay recorded audio and inspect probability values.

---

Recording Tool

Generate clean WAV recordings for model evaluation.

---

Server Integration

Test production server communication before deployment.

---

🚀 Installation

Clone the repository

git clone https://github.com/dhruvpandit46/Model-Test.git

Go inside the project

cd Model-Test

Run

Simply open

index.html

or

record.html

or

test_offline.html

using a local web server.

---

🎯 Future Improvements

• Multiple wake words
• Noise robustness evaluation
• ROC Curve visualization
• Accuracy benchmarking
• False positive analysis
• TensorFlow Lite support
• WebGPU acceleration
• Mobile optimization
• Confidence heatmaps
• Automatic dataset generation

---

🤝 Contributing

Contributions are welcome.

1. Fork the repository

2. Create your feature branch

3. Commit your changes

4. Push your branch

5. Open a Pull Request

---

📜 License

Licensed under the MIT License.

MIT © 2026 Dhruv Pandit.

See the LICENSE file for full license details.

---

👨‍💻 Author

Dhruv Pandit

GitHub

https://github.com/dhruvpandit46

LinkedIn

https://linkedin.com/in/dhruv-pandit-755786326

Instagram

https://instagram.com/dhruv_pandit2007

---

🌌 Aniquen Ecosystem

Repository	Description
🤖 Aniquen	AI Voice Assistant
🧠 aniquen-server	Backend AI Inference Engine
🧪 Model Test	Wake Word Research & Testing Environment
🎤 Anique AI Voice	Text-to-Speech Server
💬 Anique AI	Browser-Based AI Chatbot

---

⭐ Support

If you found this project useful,

please consider giving it a ⭐ on GitHub.

It helps support future development.