Skip to content

ASR4/skyparse

Repository files navigation

SKYPARSE

Intelligent Air Traffic Control Voice & Situational Awareness Platform

SKYPARSE is an AI system that transcribes live ATC (Air Traffic Control) radio communications, extracts structured intent—callsigns, instructions, and parameters—and maps them to real-time aircraft positions using ADS-B surveillance data. It serves as the foundational intelligence layer for air traffic control AI, starting with voice comprehension and expanding into decision support, compliance monitoring, and safety analytics.


Architecture

┌──────────────────────────────────────────────────────────────┐
│                         SKYPARSE                             │
│                                                              │
│  ┌──────────┐   ┌──────────────┐   ┌──────────────────────┐ │
│  │  Audio    │──▶│  ASR Engine  │──▶│  Intent Parser       │ │
│  │  Ingestion│   │  (Whisper    │   │  (NLP / LLM)         │ │
│  │           │   │   fine-tuned)│   │                      │ │
│  └──────────┘   └──────────────┘   └──────────┬───────────┘ │
│                                                │             │
│  ┌──────────┐                       ┌──────────▼───────────┐ │
│  │  ADS-B   │──────────────────────▶│  Fusion Engine       │ │
│  │  Feed    │                       │  (Match intent       │ │
│  │          │                       │   to aircraft)       │ │
│  └──────────┘                       └──────────┬───────────┘ │
│                                                │             │
│                                     ┌──────────▼───────────┐ │
│                                     │  Visualization       │ │
│                                     │  (Web Dashboard)     │ │
│                                     └──────────────────────┘ │
└──────────────────────────────────────────────────────────────┘

Data flow: LiveATC audio → noise reduction & VAD segmentation → Whisper ASR transcription → LLM-based intent extraction → callsign-to-aircraft fusion with ADS-B → real-time dashboard visualization.


Quick Start

Using Docker Compose (recommended)

# 1. Clone and configure
cd skyparse
cp .env.example .env
# Edit .env with your API keys (ANTHROPIC_API_KEY, etc.)

# 2. Launch all services
docker-compose up --build

# 3. Access
#    API:       http://localhost:8000
#    Dashboard: http://localhost:3000

Manual Setup

# 1. Create a virtual environment
python -m venv venv
source venv/bin/activate  # or `venv\Scripts\activate` on Windows

# 2. Install dependencies
pip install -e ".[dev]"

# 3. Start PostgreSQL + TimescaleDB (requires Docker)
docker-compose up -d db

# 4. Configure environment
cp .env.example .env
# Edit .env with your credentials

# 5. Run the API server
uvicorn api.main:app --reload --host 0.0.0.0 --port 8000

Tech Stack

Layer Technology
ASR Model Whisper large-v3 + LoRA fine-tuning (PyTorch, HF PEFT)
Intent Parser Claude API (prototype) → fine-tuned Llama/Mistral (prod)
Audio Pipeline librosa, soundfile, noisereduce, Silero VAD
ADS-B Data OpenSky Network API
Backend FastAPI + WebSocket
Frontend React + TypeScript + Mapbox GL JS
Database PostgreSQL 16 + TimescaleDB
Deployment Docker, AWS (ECS/EKS)
ML Training PyTorch + Hugging Face Transformers + W&B

Project Structure

skyparse/
├── pyproject.toml              # Python project configuration
├── Dockerfile                  # API service container
├── docker-compose.yml          # Full-stack orchestration
├── .env.example                # Environment variable template
│
├── data/
│   ├── collectors/             # LiveATC downloader, OpenSky ADS-B collector
│   ├── processors/             # Audio preprocessing, transcript labeling, ADS-B processing
│   └── schemas/                # Intent taxonomy, Pydantic models
│
├── models/
│   ├── asr/                    # Whisper fine-tuning, evaluation, inference
│   ├── intent/                 # LLM parser, local parser, callsign resolver
│   └── fusion/                 # Temporal alignment, compliance tracking
│
├── api/
│   ├── main.py                 # FastAPI application entry point
│   ├── routes/                 # HTTP & WebSocket endpoints
│   └── services/               # Pipeline orchestration, database ops
│
├── frontend/                   # React + TypeScript dashboard
│   └── src/
│       ├── components/         # RadarView, TranscriptFeed, InstructionTimeline
│       ├── hooks/              # useWebSocket, useAircraftState
│       └── utils/              # Aviation math, Mapbox layers
│
├── tests/                      # pytest test suite + fixtures
├── exploration/                # Jupyter-style exploration scripts
├── scripts/                    # Dev setup, data download, pipeline runners
└── docs/                       # Architecture, ATC phraseology, deployment guides

License

MIT

About

Real-time ATC radio intelligence platform that transcribes live air traffic control audio, parses controller instructions, and fuses them with ADS-B flight data for situational awareness. Built with Python, Whisper, Claude, FastAPI, TimescaleDB, and React.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors