AI Manipulation Detector — Spotting Deceptive Speech with Machine Learning

💡 What Is AI Manipulation Detector?

AI Manipulation Detector is a web-based app (and accompanying Python backend) that analyzes audio recordings for signs of manipulative or persuasive speech patterns. Whether you’re fact-checking a podcast, auditing a sales pitch, or studying political rhetoric, this tool highlights linguistic cues—like emotional triggers or subtle framing—so you can make informed judgments.

---

🎯 Why I Built It

In an age of deepfakes and persuasive audio, we often take spoken words at face value. I wanted a way to shine a light on speech tactics—so listeners can recognize when language is being used to influence, rather than inform. This project taught me how to combine speech-to-text, NLP, and pattern analysis into a single pipeline.

---

🚀 Key Features

• 🗣️ Speech-to-Text: Converts audio files into text using Hugging Face models.

• 🔍 Pattern Detection: Scans transcripts for manipulative cues (e.g., exaggerated positivity, guilt triggers, absolutes, loaded questions).

• 📊 Visual Reports: Generates simple HTML charts showing frequency of manipulative patterns.

• ⚙️ Configurable Rules: Customize which patterns to flag or weight most heavily.

• 💾 Downloadable Results: Export your analysis as JSON or CSV for further study.

---

🧰 Tech Stack

• Backend (Python)

  • Speech-to-Text: Hugging Face’s Wav2Vec2

  • NLP & Pattern Matching: spaCy + custom regex rules

  • Web Framework: Flask

• Frontend (HTML/CSS/JS)

  • Static charts generated with Chart.js

  • Simple UI for uploading audio & viewing results

• Deployment: Self-hosted or Dockerized for local use

---

📂 How It Works (High-Level)

1. Upload Audio → 2. Transcribe → 3. Analyze Text → 4. Render Report
   Under the hood, speech is first transcribed, then passed through an NLP pipeline to detect defined manipulative patterns, and finally visualized in an easy-to-read dashboard.

---

🌱 What I Learned

• Integrating large speech-to-text models in a lightweight app

• Designing maintainable NLP pipelines with spaCy

• Balancing accuracy with performance for real-time analysis

• Building a user-friendly interface around complex ML tasks

---

🔗 Try It & Contribute

• 💻 Source Code & Instructions: 📂 GitHub Repository

• 🚀 Demo & Samples: Upload an audio file in the uploads/ folder and run app.py locally.

Contributions welcome—submit issues or PRs for new pattern rules, UI improvements, or model upgrades!

---

🤝 Let’s Connect

• 🐙 GitHub: Aarti-panchal01

• 💼 LinkedIn: Aarti-panchal01

• 📧 Email: aartipanchal539@gmail.com