Trigger Word Detection Model

This repository contains an AI-powered Trigger Word Detection system that listens for a specific keyword ("robo") in an audio stream. The model processes audio in real-time and can be integrated into voice assistants, smart devices, or hands-free control systems.

📌 Project Overview

🔹 Implemented trigger word detection from scratch using deep learning
🔹 Built a custom speech dataset with positive (trigger word) and negative samples
🔹 Used MFCC features & spectrograms for audio preprocessing
🔹 Real-time detection without saving to .wav files
🔹 Full implementation in a step-by-step Jupyter Notebook

🚀 Installation & Setup

1️⃣ Clone the Repository

git clone https://github.com/Moaz0009/Trigger-Word-Detection-Model.git
cd Trigger-Word-Detection-Model

2️⃣ Install Dependencies

pip install -r requirements.txt

3️⃣ Run Real-Time Trigger Word Detection

python main.py

🛠️ How It Works

🔹 Step 1: Data Synthesis

We generate a custom dataset by mixing trigger words, random words, and background noise.

from pydub import AudioSegment
import numpy as np

# Load background noise & speech samples
background = AudioSegment.from_wav("data_set/bk/chunk_024.wav")
speech = AudioSegment.from_wav("data_set/dd/detect (13).wav")

🔹 Step 2: Feature Extraction (MFCC & Spectrograms)

Convert raw audio into features suitable for deep learning.

from trigger_util import get_mfcc_features

features = get_mfcc_features("data_set/dd/detect (13).wav")

🔹 Step 3: Model Training

We use a deep learning model (LSTM/CNN-based) to detect the trigger word.

from keras.models import Sequential
from keras.layers import Conv1D, LSTM, Dense

model = Sequential([
    Conv1D(filters=64, kernel_size=5, activation="relu"),
    LSTM(units=128, return_sequences=True),
    Dense(units=1, activation="sigmoid")
])
model.compile(optimizer="adam", loss="binary_crossentropy", metrics=["accuracy"])

🔹 Step 4: Real-Time Inference

Instead of processing .wav files, the model directly processes raw audio arrays for low-latency detection.

def detect_trigger(audio_array):
    features = extract_features(audio_array)
    prediction = model.predict(features)
    return prediction > 0.5

🤝 Contributing

Contributions are welcome! Feel free to open issues or pull requests.

📜 License

This project is licensed under the MIT License.