🎧 Voice Advertisement Analysis with Deep Learning

📌 Overview
builds a deep learning system to analyze voice advertisements and identify what makes them effective.
It combines audio signal processing + AI to extract insights that improve digital marketing performance.

🚀 Key Highlights

🎙️ Audio classification (Ad vs Non-Ad)
🧠 Hybrid CNN + LSTM model
📊 Feature extraction (MFCC, Spectrograms)
☁️ Cloud-based pipeline using Azure
📈 Insights for marketing optimization
🧠 Model
CNN → captures audio patterns (pitch, tone)
LSTM → captures speech flow over time

➡️ Result: understands both what is said and how it is said

🛠️ Tech Stack

Python, NumPy, Pandas
Librosa (audio processing)
TensorFlow / Keras
Microsoft Azure (Blob Storage, Spark)
Matplotlib, Power BI

📂 Dataset

214K+ audio files collected
Final: 30K high-quality samples
Split: 70% Train / 15% Val / 15% Test

Preprocessing:

16kHz WAV, mono
Noise & silence removal
MFCC feature extraction

🔬 Approach

Data collection & preprocessing
Feature extraction (MFCC + spectrograms)
Model training (CNN + LSTM)
Evaluation & visualization
Business insight generation

📊 Key Results

Positive tone → +8% engagement
Good delivery → +23% sales impact
~30s ads perform best
Balanced repetition improves recall
Optimize tone, pacing, and clarity
Improve ad engagement & retention
Enable data-driven marketing decisions