 | Informative features based on linear spectral ROOT SYSTEMS IN RECOGNITION OF SPEECH TEAM (Automatic translation) In this paper we have investigated the roots of the spectral line as the primary characteristics of speech signals, the study of methods of work with dictionaries of standards in the problem of recognition of speech commands.
Ongoing studies are based on common methods of system analysis, the theory of digital signal processing, programming, methods of mathematical statistics and probability theory.
As a tool for research by the author developed a software package that allows you to analyze speech signals, working with databases, dictionaries, perform various statistical calculations and to receive tabular and graphical presentation of results.
We obtained the following new scientific results: As the primary informative features for recognizing speech commands used linear spectral background; a criterion for assessing the reliability of speech recognition commands to effectively deal with false positives recognition algorithm, proposed a method for assessing the quality standards of the generated dictionary that allows you to identify the similarities to algorithm for recognizing the team, shows how to solve the problem of finding speech units (including pauses) in a continuous flow of speech on the basis of the proposed procedure of recognition.
|
