borisdiakur/distilbert-ATC-pilot-classifier-onnx

DistilBERT ATC/Pilot Role Classifier (ONNX)

A lightweight DistilBERT-based sequence classifier fine‑tuned to distinguish between Air Traffic Controller (ATC) and Pilot utterances.
Exported to ONNX format for fast, local, offline inference — ideal for real‑time aviation communication systems and compatible with Transformers.js v5.

This model is used to automatically label transcripts from aviation radio communications, especially when paired with ASR systems such as Whisper.

---

✈️ What the model does

Given a short text snippet (typically a single radio transmission), the model predicts:

• ATC
• PILOT

It is robust to noisy, imperfect ASR output and real‑world phraseology.

---

🧠 Base model

This classifier was fine‑tuned from:

distilbert-base-uncased

---

📦 ONNX export

The model is exported using the Hugging Face Optimum ONNX exporter and organized in the folder structure expected by Transformers.js v5:

distilbert-atc-pilot-classifier-onnx
├── config.json
├── onnx
│   └── model.onnx
├── special_tokens_map.json
├── tokenizer.json
├── tokenizer_config.json
└── vocab.txt

---

🧪 Training data

This model was trained on the excellent ATC/Pilot Speaker Role Classification Dataset created by jacktol:

Dataset: https://huggingface.co/datasets/jacktol/atc-pilot-speaker-role-classification-dataset
Author: https://huggingface.co/jacktol

Huge thanks to jacktol for making this dataset publicly available.

---