A.I. Armer – Ph. D. (Eng.), Associate Professor, Department «Applied Mathematics and Informatics», Ulyanovsk State Technical University
V.S. Moshkin – Post-graduate Student, Department «Information Systems», Ulyanovsk State Technical University
This article describes a method for forming the voice command templates library using ontology in order to increase their further rec-ognition. Combinations of synonyms, each of which constitutes a new reference library are analyzed, which aims – to identify the library, which is achieved by using the highest probability of correct recognition of speech commands in the background acoustic noise from a limited vocabulary.
Of OWL-ontology allows to connect domain objects with each other, so that in the processing of any semantically meaningful expression (in this case, it is the voice command), it is possible to extract synonymous expressions needed to solve the problem of definition and content of standard-library.
To solve the problem of obtaining the maximum number of synonymous variants in the library of the voice command standards de-veloped OWL-ontology that consists of 3 classes, 89 classes of objects.
Algorithm for obtaining the list of synonyms of expression on the basis of objective OWL-ontology includes the following steps:
1. Split the input expression to semantically meaningful combinations (including the use of prepositions).
2. Search of objects obtained in the subject OWL-ontology.
3. Generate a set of expressions by changing the order of words in combination.
4. Search for synonyms for each word of the input combinations.
5. Generate a set of expressions by changing the order of words and based on their substitution by synonyms.
6. Search synonyms stable expression in the input terms (number of words cannot be the same).
7. Generation of a set of expressions by changing the order of words and synonyms fixed expressions.
8. Building a complete list of synonyms received.
For the tests taken from a library of eight voice commands of aviation problem area.
897122304 combinations were formed. Each combination was seen as a reference library of the voice command. The number of errors by identifying the utterance duration is determined for each library. The identification was made by comparing the length of recognizable with lengths of standard commands.
As a result of the experiment in the identification of eight test the voice command minimum amount of the second kind of error was 46 in the absence of the first kind of error that shows a good efficiency of the developed algorithm.
Thus, using OWL-ontology, allowing to obtain sets of synonyms for each voice instruction of the standards library possible to determine better standards library in which commands are more distinguishable by the utterance duration that is in the process of identifying the voice command using this library was the minimum number error of the second kind in the absence of errors of the first kind. This theme is not lost and the practical meaning of the library.
Galunov V.I., Solovev A.N. Sovremennye problemy v
oblasti raspoznavanija rechi / Pod red. S.V. Emeljanova // Informacionnye
tekhnologii i vychislitelnye sistemy. 2004. № 2.
Krasheninnikov V.R., Armer A.I., Sinicin I.N.
Modeli izmenchivosti rechevykh komand // Naukoemkie tekhnologii. 2007. № 9. S. 56−64.
Krasheninnikov V.R., Krasheninnikova N.A., Kuznecov V.V.
Algoritm vybora ehtalonov rechevykh komand pri raspoznavanii rechi // Trudy
Rossijjskogo nauchno-tekhnich. obshhestva radiotekhniki, ehlektroniki i svjazi im. A.S. Popova.
2007. № LXII. S. 158−159.