twangodev/tartanaviation-atc-adsb-utterances

TartanAviation ATC + ADS-B (Utterances)

Speech utterances split from twangodev/tartanaviation-atc-adsb by voice-activity detection (pyannote/segmentation-3.0). Each row is one speech segment (16 kHz mono) with the ADS-B from its parent clip.

531,050 utterances · ~398 h speech · 16 kHz mono · 67% carry ADS-B. From 40,899 of 41,823 clips (silent clips have no utterances). Built with squawk.

Usage

from datasets import load_dataset

ds = load_dataset("twangodev/tartanaviation-atc-adsb-utterances", split="train", streaming=True)
ex = next(iter(ds))
ex["audio"]          # {'array': ..., 'sampling_rate': 16000}
ex["tails"]          # aircraft callsigns, e.g. ['EJA660', 'RPA4996']
ex["clip_offset_s"]  # where the utterance starts within its clip

Schema

column	type	notes
clip_id	string	parent clip {airport}/{date}/{n}
utterance_id	string	{clip_id}/{i}
airport	string	kagc / kbtp
date	string	MM-DD-YY
start / end	timestamp	utterance window (audio and ADS-B share one clock)
duration_s	float	utterance length
clip_offset_s	float	utterance start, in seconds from the parent clip's start
n_aircraft / tails	int / list	aircraft with ADS-B (0 / [] when none)
tracks	list of struct	ADS-B pings: t, lat, lon, alt, speed, heading, tail
audio	Audio(16000)	16 kHz mono WAV

ADS-B comes from the parent clip, so utterances of the same clip share it. For per-clip ADS-B including silent clips, see the clip dataset.

License

CC-BY-4.0, inherited from CMU AirLab's TartanAviation. Credit CMU AirLab and cite:

@article{patrikar2024tartanaviation,
  title={TartanAviation: Image, Speech, and ADS-B Trajectory Datasets for Terminal Airspace Operations},
  author={Jay Patrikar and Joao Dantas and Brady Moon and Milad Hamidi and Sourish Ghosh and Nikhil Keetha and Ian Higgins and Atharva Chandak and Takashi Yoneyama and Sebastian Scherer},
  year={2024},
  eprint={2403.03372},
  archivePrefix={arXiv},
  primaryClass={cs.LG},
  url={https://arxiv.org/pdf/2403.03372.pdf}
}