Periodic Reporting for period 2 - ATCO2 (Automatic collection and processing of voice data from air-traffic communications)
Période du rapport: 2020-11-01 au 2022-02-28
ATCO2 project aimed at at developing a unique platform allowing to collect, organize and pre-process air-traffic control (voice communication) data from air space. The built platform can operate over the long-term and allows various kinds of bodies to access the data to be used to develop different kind of AI (machine learning) applications, specifically those related to automatic recognition of voice recordings of air-traffic controllers.
As voice is the most natural way of communication, it is currently recognized as a priority alternative in near-future human-machine interaction. Human-machine interfaces using a comprehensive speech recognizer are proven to be highly efficient so it is expected to provide benefits in any currently foreseeable technological environment in ATM.
Overall objectives of the project were (i) to develop a sustainable platform, (ii) collect large volume of real-time and offline data from VHF channels, (iii) implement and integrate state-of-the-art machine learning technologies supporting active learning, (iv) expand community of contributors and (v) fully support legal and ethical compliance.
All the objectives of the project were achieved. The project collected large quantity of speech data, further automatically pre-processed and aligned with contextual information available from ADS-B sources. The data is offered for research as well as for commercial applications. The project platform is aligned with an already existing OpenSky Network.
As voice is the most natural way of communication, it is currently recognized as a priority alternative in near-future human-machine interaction. Human-machine interfaces using a comprehensive speech recognizer are proven to be highly efficient so it is expected to provide benefits in any currently foreseeable technological environment in ATM.
Overall objectives of the project were (i) to develop a sustainable platform, (ii) collect large volume of real-time and offline data from VHF channels, (iii) implement and integrate state-of-the-art machine learning technologies supporting active learning, (iv) expand community of contributors and (v) fully support legal and ethical compliance.
All the objectives of the project were achieved. The project collected large quantity of speech data, further automatically pre-processed and aligned with contextual information available from ADS-B sources. The data is offered for research as well as for commercial applications. The project platform is aligned with an already existing OpenSky Network.
More specifically, following objectives were achieved:
The project partners have finalised the VHF receiver, objectively evaluated through several measures.
The project partners have developed a back-end platform allowing to transfer the collected data from the receivers operated by the data feeders.
The project platform allows to exploit the VHF data (i.e. data available from different sources). Further, our own VHF data were collected in the project through our own receivers.
The collected data allowed to train and evaluate several versions of machine learning models (including automatic speech recognition, callsign detection, automatic alignment of the VHF recordings with the ADS-B data available at the OpenSkyNetwork cloud, automatic extraction of concepts from text, etc.)
Set of manually verified and automatically pre-processed voice recordings was released and the project contacted a group of active contributors.
Several studies related to legal compliance and especially processing of personal data were performed.
The project partners have finalised the VHF receiver, objectively evaluated through several measures.
The project partners have developed a back-end platform allowing to transfer the collected data from the receivers operated by the data feeders.
The project platform allows to exploit the VHF data (i.e. data available from different sources). Further, our own VHF data were collected in the project through our own receivers.
The collected data allowed to train and evaluate several versions of machine learning models (including automatic speech recognition, callsign detection, automatic alignment of the VHF recordings with the ADS-B data available at the OpenSkyNetwork cloud, automatic extraction of concepts from text, etc.)
Set of manually verified and automatically pre-processed voice recordings was released and the project contacted a group of active contributors.
Several studies related to legal compliance and especially processing of personal data were performed.
Large improvements were achieved in project performing automatic speech recognition of air-traffic communication. This specifically also includes automatic detection of callsign, training using automatically generated labels, and recognising concepts from the recognized text. Further, several natural language processing algorithms were proposed, implemented and integrated in the final project platform (including speaker role detection, or word tagging.
The data collected in the project and the developed platform allow in a wider-context to:
Exploit this data by expert to train various AI components for ATM,
Integrate the platform including automatic speech recognition systems into the real-world applications (i.e. ATM).
The data collected in the project and the developed platform allow in a wider-context to:
Exploit this data by expert to train various AI components for ATM,
Integrate the platform including automatic speech recognition systems into the real-world applications (i.e. ATM).