EML Speech Technology is Partner in the Research Project „AdaptAR“

Manuals that know more: Better and more cost-effective product documentation thanks to augmented reality and digital twins Everyone is familiar with the situation: Instructions and manuals that do not meet the expectations and needs of their users. Either important information is missing or it was too late for subsequent changes to be taken into account in the printed manual. In addition, customers need information prepared differently depending on their technical knowledge. As a result, complex manuals are a horror to many users as they are not able find the right answers to their specific questions.

In the BMBF (Federal Ministry of Education and Research) funded project "Adaptive and automatic generation of instructions in an AR environment" - in short: "AdaptAR" - a consortium of twelve partners under the leadership of the Fraunhofer Institute for Production Technology IPT is developing an augmented reality system that will reduce the cost to create technical instructions by approx. 70 percent and make instructions more user-friendly.

»So far our service technicians often have to travel a long way to get to the site. Customized instructions and remote service offerings would save a lot of time", describes Pia Gausemeier from project partner Miele the benefits of „AdaptAR".

For each product, a digital twin is provided that brings together all related order, product, process and resource data in a local database. This twin is enriched with additional data from connected IT systems while the product is in use. In this way, the system always includes all important information on the product and can offer the user the appropriate instructions for his product. Equipped with Smart Glasses or a tablet, they can now be guided step by step to an individual solution to their problem directly on site.

EML Speech Technology will develop and integrate a powerful component for automatic speech recognition for industrial environments. In addition to a high recognition accuracy, it will be able to provide results in real-time. The application scenarios in the project provide a technological challenge because of difficult acoustic environments, the use of special vocabularies, and required multilingualism of the application. So far, these aspects have not been sufficiently tackled in commercially available speech recognition systems and will be addressed as follwos:

• Development of online-capable algorithms for robust voice activity detection, which allow an adaptation to the acoustic conditions in industrial environments (e.g. high noise level, low transmission bandwidth, non-stationary noise)
• Utilization of latest techniques for end-to-end modeling with deep neural algorithms and comparison with conventional deep learning based approaches for multilingual speech recognition
• Adaptation of vocabulary and language model of the speech recognizer to the specified use cases

Employee rights and data protection also play an important role in the project. For this reason, the partners work closely with the law firm reuschlaw Legal Consultants as well as representatives of the IG Metall trade union, who examine data protection, data security and product liability as well as ergonomics and acceptance of the new technological offerings.

Project duration is from 1.6.2020 to 31.05.2023. The research project is funded by the BMBF under the measure "Internet-based services for complex products, production processes and plants (Smart Services)".
Projekt partners:

• Fraunhofer-Institut for Production Technology IPT, Aachen
• EML Speech Technology GmbH, Heidelberg
• Werkzeugmaschinenlabor WZL (of RWTH University), Aachen
• Aixtron SE, Herzogenrath
• Fionec GmbH, Aachen
• Linguatec Sprachtechnologien GmbH, Munich
• Miele & Cie. KG, Gütersloh
• Inform GmbH, Aachen
• Oculavis GmbH, Aachen
• reuschlaw Legal Consultants, Berlin
• IG Metall, Frankfurt am Main
• DigitalHUB Aachen

More information about „AdaptAR"

BMBF announcement