Welcome to the official web site of the HIWIRE (Human Input That Works In Real Environments) Project, sponsored by the European Commission. The HIWIRE project aims to make significant improvements to the robustness, naturalness, and flexibility of vocal interaction between humans and machines.

Download the Project Brochure

Download the HIWIRE database description paper (full citation in HiwirePublications)

Meet us at INTERSPEECH 2007 - special session Novel techniques for the NATO non-native Air Traffic Control and HIWIRE cockpit databases

The Big Picture: HIWIRE Project Overview

Despite good progress over the last few decades, speech recognition technology is still not sufficiently reliable for use in many real-world applications. The HIWIRE project aims to make significant improvements to the robustness, naturalness, and flexibility of vocal interaction between humans and machines. The industrial partners involved in the project expect that the technological breakthrough targeted in this project will impact their future activities by:

• Enabling the introduction of vocal dialogue with equipment in commercial aircraft cockpits.

• Improving the potential for vocal interaction with PDAs and other mobile devices in aeronautic application environments.

It is intended that the HIWIRE project will go well beyond the current state-of-the-art for robust vocal dialogue involving large flexible vocabularies, by focusing on two main targets: improved robustness against the environment (mostly unpredictable noises like those encountered in cockpits with dense audio traffic or factory noise) and improved tolerance to user behaviour (including speakers' vocal individuality, different accents, non-native speech, dialogue skills, etc). In order to guarantee common progress toward these two targets, integration nodes have been included in the project work plan, where the convergence toward all objectives will be verified. Performance will be evaluated on standard databases (for benchmarking) or on specifically collected ones for more specific issues, such as non-native speech dialogue. The project includes five milestones:

• M1 m10 Completion of baseline experiments on robustness.

• M2 m21 Completion of phase1 experiments on robustness.

• M3 m24 Completion of the integration of a first version of the advanced spoken dialogue system into the application platforms.

• M4 m33 Completion of phase2 experiments on robustness.

• M5 m33 Completion of evaluation on platforms.

Project Objectives

One clear trend in the global evolution of human society is the growing role played by information and communication. In the near future, the proliferation of new technologies and services that will become available to European citizens has the potential to offer real benefits to each individual and is expected to improve the quality of life of the entire population. This evolution underlines the need for "intuative" easy to use interfaces that are able to hide increasing levels of complexity in the underlying technology. An ideal solution would be that machines could "understand" human users through interfaces based on natural modalities, allowing the human to establish a convenient and generic dialogue with their surroundings. Also, by analogy with human-to-human communication, it is expected that the main modality of interaction between human and machine shall be speech, as it is the most natural for communication purposes. However, the introduction of natural interaction between human and machines requires powerful, and more particularly, robust automatic speech understanding. Unfortunately, the capabilities of spoken dialogue systems in fixed and mobile embedded applications have not improved as much as expected; both application developers and users still have to struggle with a significant lack of dependability in real-world applications.

The key limitation of current systems is an unreliable level of performance as a function of the noise conditions in the application environement and the individuality of user behaviour. For example, performance can be catastrophic for non-native speakers. As a consequence, human-machine dialogue based on the speech modality is not convenient today in an open environment (such as an aeroplane, a factory, an urban street, etc.) and is still at an incipient stage of development. This lack of maturity is directly perceptible at the market level for various types of applications. For example, in aeronautics the lack of dependability prevents the introduction of spoken dialogue systems into:

• fixed installations such as an aircraft cockpit

• mobile equipment such as ruggedised PCs or PDAs for aircraft maintenance.

Both of these application areas raise strong interest in the aeronautic community as speech-based interaction could provide a major contribution to enhanced safety and efficiency. These examples demonstrate that there is an across-the-board need for performance improvement. In order to create the conditions to settle an acceptable and robust natural interaction between human and machines, it is necessary to create a breakthrough in performance for robust speech understanding. It is therefore assumed that the following requirements are of paramount importance:

• Flexible vocabulary up to several thousands of words.

• Accessible by all users whatever their accent (including non-native speakers).

• Insensitive to the user environment (noisy vehicle or aircraft, conversational noise, and noisy external environment).

• Advanced dialogue capability in order to interpret the behaviour ofa range of skilled and unskilled users.

The overall objective of the HIWIRE project is to set the basis for much more dependable speech recognition in mobile, open and noisy environments, and provoke the necessary technical breakthroughs. The achievements of the project will be validated through:

• Assessment of the potential of contribution of vocal interaction to safety and efficiency in future commercial cockpits.

• Usability evaluation of enhanced dialogue in an open environment on a mobile device.

This main objective at a strategic level is split into three working objectives:

_Objective 1 : To make significant improvements to the robustness of speech recognition in noisy environments._

_Objective 2 : To make significant improvements to the robustness of speech recognition to different users' voices and_ _interaction abilities._

_Objective 3 : To evaluate the potential impact of more robust speech recognition in real-world applications._

---

Notes:

• You are currently in the HIWIRE web. The color code for this web is this background, so you know where you are.
• If you are not familiar with the TWiki collaboration platform, please visit WelcomeGuest first.

• HIWIRE is a STREP (Specific Targeted REsearch or innovation Project) funded by the EC 6th Framework IST Programme IST Programme