Speech technology

In the last years ALL begun to use its research results in the field of speech- and language technologies. The earlier research and development experiences and the new approaches brought big success also in this field. As a result of the speech technological researches ALL worked out a unique technology for indexing and searching Hungarian language loud storages and at present time we are close to the practical application of continuous speech recognizing. We produced the prototype of a semantic searching system during our language technological researches. With the semantic searcher we can realize subject based searching, instead of usual word based searching.

As to the effect of efficient spoken or written communication with humans a special natural language dialogue technology will be developed. This technology is built upon a special knowledge based natural language understanding method. Understanding is supported by domain ontology and the formal description of the natural language grammar. Domain ontology defines the concepts and relationships included in the universe of discourse, which allows knowledge-based approach to be used. The domain ontology will be maintained continuously by the use of specific intelligent tools. Therefore the technology provides developing and adaptive ontology and semantics representation. Using the so-called construction networks, which are more suitable to describe different natural languages than the usual generative grammars, specifies natural language grammar. Domain ontology and grammar together allow spoken or written queries and documents to be converted into semantic representations, which then are matched to each other by various reasoning techniques such as abduction, deduction and analogies.

Multilingual Speech Recognition System

The Speech Recognizing System of ALL is a standalone software product that is prepared for converting digitally recorded speeches to the corresponding sequences of words. The domain of the recognizable speeches is practically non-constrained, as the software is enabled to manage more than 300 000 spoken words.

Retrieval of Spoken Information from Large Audiovisual Archives

The Speech Retrieval product of ALL offers the possibility to search through large audiovisual archives for speech just in the same manner as the search for texts can be realized in usual information retrieval systems.
The system encompasses two self-contained modules: Preprocessing and Retrieval.

Vocal Query Interface

Vocal Query Interface is a peculiar speech recognizing product of ALL, which enables manually directed electronic systems to be controlled by voice-initiated commands (words and phrases).

Speaker Recognition System