GPT Lie Analysis

An automated AI pipeline for speech-based lie / scam risk analysis, combining Audio preprocessing → Speech-to-text (Whisper) → LLM reasoning → Alerts.

This project is designed for hands-free background monitoring (e.g. mobile call recordings) and supports automatic Gmail / Discord notifications when high-risk speech patterns are detected.

---

✨ Features

• 🎧 Audio & video input support (.wav, .mp3, .m4a, .mp4, etc.)
• 🧠 AI-based lie / scam risk analysis via LLM
• 📊 Risk scoring with configurable threshold
• 📬 Automatic Gmail alert
• 💬 Discord webhook notification
• 🔁 Folder monitoring (polling-based, stable & mobile-friendly)
• 🛡 Guardrails to avoid false alerts on very short / insufficient audio
• 📱 Designed to run persistently on Termux (Android) or Linux

---

📦 Project Structure

gpt_lie_analysis/
├── src/gpt_lie_analysis/
│   ├── core.py                 # Main pipeline logic
│   ├── whisper_transcribe_api.py
│   ├── gpt_lie_analysis_requests.py
│   ├── audio_loader.py
│   ├── segment_and_analyze.py
│   ├── gmail_notify.py
│   └── cli/
│       └── app.py              # CLI entrypoint
├── scripts/                    # Helper scripts (optional)
├── tests/
├── pyproject.toml
├── requirements.txt
└── README.md

---

🔧 Requirements

• Python ≥ 3.10
• ffmpeg
• OpenAI API Key

Install system dependency

sudo apt install -y ffmpeg

---

🔑 Environment Setup

Create a .env file in project root:

OPENAI_API_KEY=sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Load it into current shell (important for Termux / nohup):

set -a
source .env
set +a

---

📥 Installation

python -m venv .venv
source .venv/bin/activate

pip install -e .

Verify:

lie-analyzer --help

---

▶️ Usage

1) Analyze a single file (manual test)

lie-analyzer one ./test.wav --threshold 7

Used mainly for:

• Debugging
• Demo recordings
• Development testing

---

2) Continuous folder monitoring (recommended)

lie-analyzer scan \
  --interval 3 \
  --call-rec "/sdcard/MIUI/sound_recorder/call_rec" \
  --recordings "/sdcard/Recordings" \
  --meta-ai "/sdcard/Download/Meta AI" \
  --threshold 7

This mode:

• Continuously monitors predefined folders
• Automatically detects new recordings
• Waits for file write completion (prevents half-written files)
• Triggers analysis + alerts when needed

---

📱 Termux Background Running

Recommended for mobile usage.

Start background monitor

nohup lie-analyzer scan \
  --interval 3 \
  --call-rec "/sdcard/MIUI/sound_recorder/call_rec" \
  --recordings "/sdcard/Recordings" \
  --meta-ai "/sdcard/Download/Meta AI" \
  --threshold 7 \
  > lie_analyzer.log 2>&1 &

echo "PID=$!"

View live log

tail -f lie_analyzer.log

Stop

pkill -f "lie-analyzer scan"

---

⚠️ Notes

• If your audio is too short, the LLM may refuse to give a score. Provide longer speech.
• For .mp4 videos, the pipeline will extract audio first (ffmpeg required).

---

License

MIT (or your preferred license)