Automation Philosophy
SELF-MADE-ATC directly addresses the significant challenges in maintaining and customizing Automatic Speech Recognition and Understanding (ASRU) in Air Traffic Control (ATC). Building on a recent technology transfer to the industry, our solution introduces a fully automated, continuous process for ASRU maintenance and adaptation.
This automation is achieved through four key features :
Core Components Explained
DATA-COLLECTION
Speech recognition models require high-quality training data specific to the application domain. While a foundational model trained on extensive ATC recordings is helpful, fine-tuning requires data from the end-user’s real operational area.
Key data preparation challenges are automatically handled by SELF-MADE-ATC:
- Transmission Segmentation: If Push-To-Talk (PTT) markers are unavailable, the system automatically segments recordings, managing both undersplitting (multiple speakers in one segment, e.g., ATCO and Pilot) and oversplitting (a single transmission broken into two files due to long pauses).
- Speaker Classification: Automatic assignment of speaker labels (e.g., ATCO vs. Pilot) to transmissions.
After recording, splitting, and classification, the system automatically transcribes and improves the data, readying it for training.
🔒 Data Privacy: All recording, splitting, speaker classification, and transcription processes are performed entirely at the end-user’s facilities. No data transfer to DLR or external ATM system suppliers is needed, fully adhering to GDPR and ensuring data privacy remains under the user’s control.
MODEL-TRAIN
The automated pipeline uses the newly transcribed data, combined with other available resources, to fine-tune a basic or existing moderate recognition model. The latter approach is preferred for its greater speed and efficiency.
🔒 Data Security: Training and fine-tuning are executed exclusively at the end-user’s site. Once the fine-tuning is complete, the utilized data can be securely and permanently deleted.
ADAPT
The high cost of adapting ASRU systems to new customers (new airports, new regions) is often a major roadblock for ATM manufacturers. SELF-MADE-ATC eliminates labor-intensive manual processes, offering adaptation cost reductions by a factor of 3 to 7 times compared to typical license fees. This significantly reduces the financial and operational burden.
Dedicated tools support adaptation for diverse applications and environments, such as:
- Radar label maintenance for approach control.
- Readback error detection on the apron.
- Adaptation between different airports (e.g., transitioning from Vienna approach to Zurich approach).
MAINTAIN
Beyond initial setup, regular maintenance costs pose another significant challenge. SELF-MADE-ATC addresses the need for continual updates—for instance, when new airlines or aircraft types are introduced, or when waypoints (like “DEXON”) are decommissioned and replaced. This system automates the update cycles, offering the same dramatic cost reduction (3 to 7 times) as the adaptation process.
MONITOR
This component provides crucial oversight by continuously tracking the ASRU system’s operational health and performance in real-time. It monitors key metrics and system logs to quickly identify and flag any degradation or deviation from expected performance standards, triggering necessary maintenance or adaptation cycles automatically.
EVALUATE
The system performs automated, rigorous quality assurance. It evaluates the accuracy and reliability of the ASRU output against predefined performance metrics, ensuring compliance with safety-critical aviation standards. This continuous evaluation loop is essential for validating the output of the Self-Maintain and Self-Adapt processes.