---
dataset_info:
  features:
  - name: id
    dtype: string
  - name: class
    dtype: string
  - name: text
    dtype: string
  splits:
  - name: train
    num_bytes: 518320.74405107374
    num_examples: 5512
  - name: validation
    num_bytes: 64884.12797446315
    num_examples: 690
  - name: test
    num_bytes: 64884.12797446315
    num_examples: 690
  download_size: 349738
  dataset_size: 648089.0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
---

# ATC-Pilot Speaker Role Classification Dataset

This dataset is designed for binary text classification in air traffic communication, distinguishing between utterances made by **pilots** and those made by **air traffic controllers (ATC)**.

## Dataset Origin

The data was derived from the [UWB ATC Corpus](https://lindat.mff.cuni.cz/repository/xmlui/handle/11858/00-097C-0000-0001-CCA1-0), a publicly available dataset of air traffic control communications provided by the University of West Bohemia. This corpus consists of approximately 20 hours of manually transcribed recordings from Czech airspace, capturing a wide range of heavily accented English and common linguistic artifacts found in international ATC exchanges.

The original transcriptions were cleaned, normalized, and labeled to support the speaker role classification task.

## Preprocessing

The dataset underwent extensive preprocessing to adapt the raw UWB transcripts for classification. Key steps included:

- **Uppercasing** all transcripts for uniformity
- **Converting digits to words** (e.g., `3 5 0 → THREE FIVE ZERO`)
- **Expanding letters to phonetic alphabet equivalents** (e.g., `N → NOVEMBER`)
- **Removing non-English**, unintelligible, or corrupted segments
- **Normalizing diacritics** and fixing broken Unicode characters
- **Manual review** of ambiguous speaker segments to ensure correct labeling

## Structure

Each example contains:
- `id`: A unique sample identifier
- `text`: The processed communication utterance (uppercase, normalized)
- `class`: The speaker role label (`"pilot"` or `"atc"`)

The dataset is fully balanced, with an equal number of `pilot` and `atc` examples in the `train`, `validation`, and `test` splits.

## Use Cases

This dataset is intended for:
- Training and evaluating speaker role classifiers in air traffic communications
- Text-based analysis in low-resource, safety-critical domains
- Preprocessing or filtering stages in multi-modal ATC systems

## Reproducibility

The full preprocessing and construction pipeline is available in the associated GitHub repository. It includes tools for raw transcript processing, dataset construction, and evaluation, allowing others to reproduce or extend the dataset as needed.

## Conclusion

This dataset supports robust and accurate classification of speaker roles in air traffic communication. It enables text-only modeling of a task that typically relies on acoustic input, and provides a realistic benchmark for structured aviation dialogue understanding. The inclusion of accented English, phonetic variability, and transcription inconsistencies makes it especially valuable for research in real-world ATC scenarios.

## References

- [UWB ATC Corpus – LINDAT/CLARIAH-CZ Repository](https://lindat.mff.cuni.cz/repository/xmlui/handle/11858/00-097C-0000-0001-CCA1-0)
- [GitHub Repository – ATC Pilot Speaker Role Classification Task](https://github.com/jack-tol/atc-pilot-speaker-role-classification-task)