Loading...
Projects / Programmes source: ARIS

OptiLEX: Research and optimization of lexical resources for embedded speech technology implementations

Research activity

Code Science Field Subfield
2.06.00  Engineering sciences and technologies  Systems and cybernetics   

Code Science Field
P176  Natural sciences and mathematics  Artificial intelligence 

Code Science Field
5.02  Social Sciences  Economics and Business 
Keywords
language resources, speech technologies, man-machine communication, embedded systems
Evaluation (rules)
source: COBISS
Researchers (16)
no. Code Name and surname Research area Role Period No. of publicationsNo. of publications
1.  23505  PhD Kozma Ahačič  Linguistics  Researcher  2018 - 2021  665 
2.  36709  Igor Bizjak  Metrology  Researcher  2019 - 2021 
3.  11805  PhD Simon Dobrišek  Computer science and informatics  Researcher  2018 - 2021  283 
4.  13144  PhD Stanislav Gruden  Electronic components and technologies  Researcher  2019 - 2020  36 
5.  28160  PhD Peter Holozan  Computer science and informatics  Researcher  2018 - 2021  58 
6.  53879  Marija Ivanovska  Systems and cybernetics  Researcher  2020 - 2021  28 
7.  31985  PhD Janez Križaj  Systems and cybernetics  Researcher  2018 - 2021  38 
8.  34592  PhD Tanja Mirtič  Linguistics  Researcher  2018 - 2021  96 
9.  36688  Klemen Pečnik  Telecommunications  Researcher  2019 - 2020  34 
10.  28792  Miroslav Romih  Computer science and informatics  Researcher  2018 - 2021  32 
11.  24772  MSc Simon Rozman  Computer science and informatics  Researcher  2018 - 2021 
12.  39512  Robert Sedevčič    Technical associate  2019 - 2020 
13.  39513  Simona Škrlj    Technical associate  2018 
14.  28458  PhD Vitomir Štruc  Systems and cybernetics  Researcher  2018 - 2021  351 
15.  11380  PhD Mario Žganec  Metrology  Researcher  2018 - 2021  98 
16.  12000  PhD Jerneja Žganec Gros  Computer science and informatics  Head  2018 - 2021  290 
Organisations (4)
no. Code Research organisation City Registration number No. of publicationsNo. of publications
1.  0618  Research Centre of the Slovenian Academy of Sciences and Arts  Ljubljana  5105498000  62,722 
2.  1538  University of Ljubljana, Faculty of Electrical Engineering  Ljubljana  1626965  27,585 
3.  1986  ALPINEON R & D  Ljubljana  1820931  387 
4.  8333  AMEBIS programska oprema, d.o.o. Kamnik (Slovene)  Kamnik  5512930  84 
Abstract
Although there are many studies and a great deal of literature addressing computational representations of lexical resources, their implementation in comprehensive speech recognition or synthesis systems on embedded platforms poses a substantial problem, which is further aggravated by limitations in the hardware used. An overview of the state-of-the-art shows that in developing speech technologies for languages with a large inflectional paradigm the use of procedures developed for English is ineffective. Due to a large number of inflections given the same number of lexemes, the scope of the Slovenian speech recognizer is at least ten times larger. This is why special attention should be directed precisely to optimizing the models used and adapting them to the morphological specifics of inflection-rich languages. Slovenian is a Slavic language that has very complex accentuation and pronunciation characteristics. The accentuation not only depends on a lexical item or individual syllable: it changes in terms of both quality and position throughout the inflectional paradigm. Speakers of Slovenian have to learn lexical stress positions along with learning the language. As a consequence, a pronunciation lexicon that indicates lexical stress positions and pronunciation variants for as many Slovenian words as possible is crucial for the development of speech technology applications and linguistic research. Such a lexicon can be used either in its full-blown form or as a training material for machine learning techniques aimed at automatically predicting word pronunciations. In the OptiLEX applied project, the focus will be on developing a reference pronunciation dictionary and on research and implementation of a module that will enable effective presentation of lexical resources and be able to function in speech engines on embedded platforms. A number of problems will be addressed: the need to operate in real time; the need to store language resources in a compact manner; and finally, the need for a small language-resource footprint in random-access memory. The objective of the OptiLEX project is to develop a reference pronunciation dictionary and to study the procedures for optimizing the computer presentation of lexical resources for inflection-rich languages that may be used for speech recognition and synthesis on embedded platforms. The procedures will be implemented and validated on the hardware of an embedded mobile terminal. The project group will seek effective procedures for reducing redundancy in the presentation and computerized storage of lexical resources for inflection-rich language groups, which will enable fast and high-quality conversion of grapheme-based entry of words into a phonetic transcription and vice versa by utilizing as little memory as possible. The effectiveness of reducing the redundancy of lexical resources will be demonstrated and validated using lexical resources for several inflection-rich languages, including Slovenian.
Significance for science
The global impact that OptiLEX will produce at the scientific and technological levels can be summarized in the following terms: · OptiLEX will contribute to progress beyond the state-of-the-art in speech technologies by advancing the methods of lexical resource representation with an emphasis on inflection-rich languages. · OptiLEX will address operational challenges in rapid implementation of speech technologies using large pronunciation lexicons on embedded mobile platforms and Internet of Things (IoT) devices with limited computational and memory resources. · OptiLEX will explore new alternatives in embedded speech technology implementations, considering alternative lexical resource representations. · OptiLEX will reuse the available language infrastructure, such as existing language resources and the existing FRAN language portal tool, and tools that were developed in part under national funding programs, taking advantage of this and exploiting synergies between new research activities and research funding already granted. · OptiLEX will contribute to generating new knowledge in linguistics by concentrating on linguistic studies of the spoken language. Research on spoken Slovenian has largely been neglected, even though the spoken language is a primary communication channel that is increasingly gaining in visibility and importance through the rapid advancement of modern communication technologies. OptiLEX results will be useful for researchers working on (standard) spoken Slovenian, and will provide an important foundation for designing a contemporary version of the normative guide for pronunciation. Namely, the most recent and only pronunciation guide was compiled seventy years ago! This means that OptiLEX is not only tackling research problems that are valid and fascinating, but is also addressing essential basic questions that Slovenian linguistics still urgently needs to answer. · OptiLEX will contribute to strengthening the European innovation capacity by committing to open knowledge and reproducible research as one of the keys for advancing the research community. The lexical representation modules implementing techniques developed in OptiLEX will be made available using free APIs. The terms of access to the pronunciation dictionary modules will be determined by the access right conditions to the primary source. For sources with open access, pronunciation dictionaries will be provided on an open access basis. All openly available modules of the reference pronunciation dictionary will be disseminated both via the FRAN portal and the language resource repository of the research infrastructure CLARIN.SI as part of the larger EU CLARIN infrastructure. This will be an important contribution to the open data initiative and will extend the usage of language resources developed in the project to the wider community. The publication of the language resource modules will not be tied to the project completion, as they will be continuously further developed also after the end of the project.
Significance for the country
OptiLEX includes exploitation planning activities in which technology transfer plays a crucial role. The challenge here is to strengthen Europe's industrial capabilities and business prospects in the language technology market by addressing the demands of the Digital Single Market. The prerequisite for the success of the Digital Single Market is that products and information be available in a multitude of languages spoken by customers and consumers. The research methodology proposed in OptiLEX, based on common standardized lexical resources and open APIs, will increase the opportunities for such a technology transfer. The global impact that OptiLEX will produce can be summarized in the following terms: · Intelligent key enabling technologies (KETs ) for multiple industrial sectors OptiLEX will advance the state-of-the-art of language technology resource representation for mobile and IoT applications and make it possible to provide Key Enabling Technologies (KETs) with a strong industrial dimension. These ICT KETs play an important horizontal role in several economic sectors linked to European Smart Specialization Strategy S4 by providing intelligent user interface technologies for the following vertical market segments: smart cities and communities, smart homes, smart mobility, factories of the future, and e-health. · Impact in the speech technologies industrial sector The global market for speech technologies is currently dominated by a few key players; however, several new opportunities exist. The current key players (such as Nuance Communications, Agnitio, and Microsoft) are currently mainly focused on large-scale implementation applications (the B2B market) and predominantly target languages with a large number of speakers. New exploitation routes have been identified, particularly in the B2C market, the mobile market, and the IoT markets under exploration by SMEs and start-ups. · Impact on the education and public administration sector OptiLEX will contribute to intelligent user interface technologies that various stakeholders use to access information and services provided by public administration on several levels. OptiLEX technologies can be used for voice-enabled access via mobile platforms to information and services as part of the e-administration project. They can further be used for automatically building e-learning materials and tools that can be of essential importance to users with reading difficulties (e.g., dyslexia) or to blind and visually impaired users. · Impact on the protection of national cultural heritage OptiLEX results will create important language resources in form of pronunciation dictionaries for both standard and nonstandard Slovenian that will capture pronunciation information on various levels thereby contributing to the preservation and protection of national cultural heritage.
Most important scientific results Interim report
Most important socioeconomically and culturally relevant results Interim report
Views history
Favourite