Datasets:
metadata
language:
- en
license: other
task_categories:
- automatic-speech-recognition
tags:
- air-traffic-control
- aviation
- atc
- pilot
- controller
configs:
- config_name: S2R
data_splits:
- name: train
num_bytes: null
num_examples: 1443
size_categories:
- 1K<n<10K
ATC Pilot–Controller Conversations
Dataset Description
This dataset contains English conversations between pilots and air traffic controllers (ATC), transcribed from real VHF radio transmissions recorded at controlled aerodromes.
Each example consists of:
- A WAV audio file (mono, VHF radio quality)
- A transcription of the utterance
- Speaker metadata: whether the speaker is the pilot or the controller
- Utterance type: e.g. readback, taxi, pushback, etc.
- Token-level annotations: callsigns, taxiways, frequencies, actions, and greetings tagged inline
- Duration of the audio clip in seconds
Configuration & Splits
| Config | Split | Description |
|---|---|---|
| S2R | train | All labelled pilot–controller utterances |
Data Fields
| Field | Type | Description |
|---|---|---|
audio |
Audio | WAV audio file |
transcription |
string | Verbatim transcript |
speaker |
string | pilot or controller |
utterance_type |
string | High-level speech act label (e.g. readback, taxi) |
metadata_raw |
string | Raw token-level annotation lines from the .txt file |
duration |
float | Audio clip duration in seconds |
conversation_id |
string | Parent conversation folder name |
file_id |
string | Stem of the source WAV filename |
Important Notes
⚠️ Audio quality — The audio is transmitted through Very High Frequency (VHF) radio.
Most recordings sound muffled or have noticeable background noise typical of cockpit and radio environments.
Models trained on clean speech may struggle without domain adaptation.
✈️ ATM-specific vocabulary — The transcriptions contain terminology specific to Air Traffic Management (ATM):
- Aircraft callsigns (e.g.
[UNK] seven zero five) which may be partially inaudible- Radiotelephony numerals: numbers are spoken in aviation phonetics
(e.g. nine → niner, five → fifer, thousand → tousand)- Taxiway designators, frequency strings, and procedural phrases
(e.g. holding point, contact tower, good day)
Citation
If you use this dataset, please acknowledge the original source of the recordings.