MikCil/f1-team-radio

F1 Team Radio Dataset

A comprehensive dataset of Formula 1 team radio communications with transcriptions.

Dataset Description

This dataset contains team radio audio clips from Formula 1 races along with their text transcriptions. Team radio communications are the real-time messages exchanged between F1 drivers and their pit wall engineers during race weekends.

Dataset Statistics

Metric	Value
Total audio clips	14,681
Grand Prix events	149
Unique drivers	43
Date range	2018-03-25 to 2025-12-07

Top Drivers by Message Count

Driver ID	Messages
LEWHAM01	1,685
MAXVER01	1,494
LANNOR01	1,137
CARSAI01	898
CHALEC01	754
GEORUS01	717
VALBOT01	686
DANRIC01	673
SERPER01	613
PIEGAS01	557

Data Fields

Field	Type	Description
id	string	Unique identifier for each radio message
driver_id	string	Driver code (e.g., MAXVER01 for Max Verstappen)
racing_number	string	Driver's car number
grand_prix	string	Full Grand Prix name (e.g., "2024 Monaco Grand Prix")
race_id	string	Race identifier (e.g., 2024_Monaco_Grand_Prix)
session_date	string	Date of the session (YYYY-MM-DD)
message_timestamp	string	UTC timestamp of the message
audio	Audio	Audio clip (MP3, resampled to 16kHz)
transcription	string	Text transcription of the radio message

Driver ID Format

Driver IDs follow the official F1 format: first 3 letters of surname + first 3 letters of first name + identifier number.

Examples:

• MAXVER01 → Max Verstappen
• LEWHAM01 → Lewis Hamilton
• CHALEC01 → Charles Leclerc
• LANNOR01 → Lando Norris

Usage

from datasets import load_dataset

# Load the dataset
ds = load_dataset("MikCil/f1-team-radio", split="train")

# View a sample
print(ds[0])

# Filter by driver
verstappen = ds.filter(lambda x: x["driver_id"] == "MAXVER01")

# Filter by race
monaco_2024 = ds.filter(lambda x: "Monaco" in x["grand_prix"])

Playing Audio

from IPython.display import Audio as IPythonAudio

sample = ds[0]
IPythonAudio(
    sample["audio"]["array"], 
    rate=sample["audio"]["sampling_rate"]
)

Fine-tuning ASR Models

This dataset can be used to fine-tune speech recognition models on F1-specific vocabulary (driver names, technical terms, etc.)

from transformers import WhisperForConditionalGeneration, WhisperProcessor

Transcription Method

Audio files were transcribed using Cohere Transcribe 03-2026, an efficient open-source automatic speech recognition model.

License

This dataset is released under the CC BY 4.0 License.

Citation

@dataset{f1_team_radio,
  author = {Michele Ciletti},
  title = {F1 Team Radio Dataset},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/MikCil/f1-team-radio}}
}

Acknowledgments

• Formula 1 for the original broadcasts
• Cohere Labs for transcription