Real-Time Trading Transcription Service

A Python-based real-time speech-to-text system designed for trading floors. This project leverages Deepgram’s Nova-3 model for accurate, low-latency transcription and OpenAI’s GPT-4.1-mini model to detect and log trade intents as they happen.

---

🚀 Overview

Trading floors are loud, fast-paced, and chaotic. Capturing accurate voice logs in real time is a major challenge, yet essential for auditing and compliance.

This project demonstrates how to:

• Stream live audio from a trader’s microphone
• Transcribe it in real time using Deepgram Nova-3
• Detect trading instructions such as "Buy 100 AAPL at market" using OpenAI’s LLM
• Log valid trades automatically into a JSONL file for record keeping

---

🧠 Architecture

The system is made up of three main components:

1. Audio Input: Captures live microphone input using PyAudio.
2. WebSocket Communication: Streams audio in chunks to Deepgram’s API for transcription.
3. Transcription & Logging: Processes transcripts with OpenAI to detect trading intents and logs detected trades in real time.

🎙️ Microphone → WebSocket → Deepgram (Nova-3) → OpenAI LLM → Trade Log

---

🧩 Features

• Real-time voice transcription
• Speaker-aware diarization
• Trading intent detection (Buy, Sell, Cancel)
• Automatic trade logging to trades_log.jsonl
• Configurable and extensible for other downstream tasks

---

⚙️ Setup

1. Clone the Repository

git clone https://github.com/Neurl-LLC/deepgram-60.git
cd deepgram-60

2. Create a Virtual Environment

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

4. Set Up Environment Variables

Create a .env file in the project root with your API keys:

OPENAI_API_KEY=sk-your-openai-key
DEEPGRAM_API_KEY=your-deepgram-key

---

🧑‍💻 Running the Application

Start the trading transcription service:

python main.py

You’ll see output like:

🎙️  Microphone live. Press Ctrl+C to stop.
🗣️  Speaker 0: Buy 100 AAPL at market
💹 TRADE DETECTED:
{
  "trade_detected": true,
  "action": "BUY",
  "symbol": "AAPL",
  "quantity": 100,
  "price_type": "MARKET",
  "price": null
}

All detected trades are stored in trades_log.jsonl with timestamps.

---

🧾 Example Trade Log

Each trade is stored as a structured JSON object:

{
  "timestamp": "2025-10-21 15:42:03",
  "trade_detected": true,
  "action": "BUY",
  "symbol": "TSLA",
  "quantity": 200,
  "price_type": "LIMIT",
  "price": 250.00,
  "raw_text": "Buy 200 Tesla at limit 250"
}

---

🛠️ Customization

You can modify the TRADING_INSTRUCTION system prompt in main.py to adapt the detection logic for different trading formats, markets, or internal compliance requirements.

---

🧩 Future Improvements

• Integrate Deepgram Flux CSR model for conversational trade execution
• Add a web dashboard for visual trade tracking
• Include keyword-based risk detection and trade validation