Skip to main content

Solutions for voice interaction towards natural crew assistant

Periodic Reporting for period 1 - VOICI (Solutions for voice interaction towards natural crew assistant)

Período documentado: 2018-03-01 hasta 2019-02-28

"VOICI aims to develop a natural crew assistant – designed to process natural spoken requests from the crew members and help them perform complex actions. Hence, it will lead to a reduction of the cognitive workload associated with the increasingly complex human-machine interfaces in modern aircraft cockpits.

VOICI's main goal is to demonstrate the natural crew assistant at Technology Readiness Level 3, i.e. experimental proof of concept. This includes speech capture via both crew headset and ambient microphone arrays, speech recognition and interpretation, and speech synthesis for dialog with the crew. An Audio Evaluation Environment is being established, where the crew assistant can be developed and tested under realistic audio/noise conditions.

The project is part of the ""extended cockpit"" work package under Clean Sky 2's ""Systems"" technology demonstration. By reducing crew workload VOICI will contribute to optimization of operations; flight safety, crew awareness, better maintenance, reduced cost of operations and generally higher efficiency and lower stress. VOICI comprises both SMEs and research institutes, and the cooperation within the consortium will contribute to innovation and job creation.

The project runs 24 months, ending Fegruary 2020. The consortium comprises 4 partners: SINTEF (NO, R&D, coordinator), sensiBel (Norway, SME), Multitel (Be, R&D), Acapela (Be, SME).
Thales (Fr, Clean Sky 2 partner) is Topic Manager.
"
During the first year, VOICI has established an Audio Evaluation Environment, emulating a Dassault Falcon 2000 cockpit. This includes a geometrical cockpit model and calibrated noise recording, split into multiple channels so as to represent multiple noise sources. Hence, a realistic noise and reverberation environment is created, including its directional properties.

A microphone array has been built, based on sensiBel's high signal-to-noise ratio optical microphone, and work on associated beamforming has started.

A first generation speech recognition framework has been set up, using Deep Neural Networks. This is independent of cloud based systems and uses dedicated language models for the cockpit scenario.

All the algorithms underlying the dialog system have been implemented and tested. From the Natural Language Understanding unit that understands natural requests to the Dialogue Core which handles the conversation flow. Particular emphasis has been placed on the ability of the voice assistant to use contextual data. All the technological components have been integrated in a single pipeline, including also crew-feedback by Acapela's Speech synthesis technology.
At completion VOICI expects to exceed the state-of-the-art by (i) improving microphone-array technology through developing methods for using high signal-to-noise ratio microphones that allow small footprint layouts, and (ii) developing embedded speech and dialog processing systems tailored to aviation language/context and background noise. These tailored components include speech recognition, natural language understanding, dialog systems and speech synthesis.
3: Embedded, standalone natural crew assistant
2: Microphone array. Flush mounted example
1: Audio evaluation environment built from sheet plywood and transparent acrylic