LUNA smooths out speech recognition
'Say Accounts if you want to speak to someone about your account'; 'Say Credit if you want to speak to someone about your credit card'. Do these statements sound familiar? If you conduct your personal business via telephones, they should. Although it saves you time and effort, automated speech recognition has its limitations. Wouldn't it be great if you could tell the service exactly what you want quickly and succinctly? The EU-funded LUNA project, funded under the Sixth Framework Programme with EUR 2.61 million, is raising the level of intelligence of automatic systems up to 'Spoken Language Understanding' (SLU). The three-year project is tackling real-time understanding of spontaneous speech in the context of advanced telecom services. LUNA's project partners, led by Ms Silvia Mosso from the Italian speech technology group Loquendo, are seeking to create a robust natural spoken language understanding toolkit for multilingual services that can carry out human-computer communication with increased user satisfaction. LUNA is tackling a number of issues by targeting language modelling for speech understanding; semantic modelling for speech understanding; automatic learning; robustness issues for SLU; and multilingual portability of SLU components. According to the researchers, SLU is more advanced than the traditional interactive voice response (IVR) systems that most people are familiar with. With IVRs, the user must respond to questions with specific words or short sentences proposed by the systems. LUNA will improve on the existing automated telephone systems by the development of menu-driven voice recognition that will result in unconstrained speech and more spontaneous interactions between humans and machines. Ultimately, LUNA will reinforce the experiences of the users. 'We had to spend a lot of time initially recording spontaneous conversations between people and between people and machines,' Ms Mosso told ICT Results. 'This is called the corpora, the collection of words and phrases that gives the software its basic language. Then, researchers have to annotate the terms in a way that machines can understand, and finally they apply statistical language models,' she explained. 'You can say things like "I have a problem with my printer" and it will help you go through the options.' The researchers believe this system will help people interact in a more natural and smoother manner. The successful applications of this project will generate faster and more productive interactions with service centres. 'The advantage with these areas is that you can apply our work to any kind of help centre,' the project coordinator said. 'But if you want to apply it to different areas, then you need to do the initial collection of the conversations, the corpora, again.' From a business angle, the researchers said LUNA's work is guaranteed practical use and France Telecom and Italy's CSI Piemonte (the Piedmont Consortium for Information Systems) are on board. From an academic/research perspective, the RWTH Aachen University in Germany, the Université d'Avignon et des Pays de Vaucluse in France, the University of Trento in Italy, the Warsaw-based Polish-Japanese Institute of Information Technology, and the Institute of Computer Science of the Polish Academy of Sciences are participating in LUNA. The researchers said the project's results will fuel competition among the industrial partners who will be able to use them directly and launch them on the speech technologies market. LUNA has already developed the most advanced SLU for the Italian and Polish languages. Ms Mosso highlighted the fact that the team will be 'refining the systems over the last months of the project', which is set to end next September.