Information Technology Reference
In-Depth Information
given speaker including extracting these features from among others that have no
major impact on current analysis results, as well as classifying the speaker. Each
of these three stages of the entire recognition process is shown in Figure 2.20.
Fig. 2.20.
Speaker recognition process
The first stage - speech signal pre-processing - consists of four basic activities
and tasks, namely the pre-emphasis, splitting into frames, eliminating superfluous
features and windowing. At this stage, two tasks mentioned at the beginning of
this chapter are executed: the speaker is verified and identified. The second stage -
significant feature selection and extracting - begins with determining the signal
strength, choosing selection coefficients, selecting extraction coefficients, and
ends in an analysis based on dynamic parameters. The last stage - speaker classi-
fication - boils down to selecting the appropriate classification methods and
classifiers.
Speech and speaker recognition are complex processes and even if they appear
independent, they have certain joint characteristics and feature some common
challenges. The latter certainly include developing a comprehensive speech recog-
nition system for our mother tongue, developing a system recognising both con-
tinuous speech and separated words, as well as designing integrated speech recog-
nition systems combined with dialogue systems.
Speech analysis and speaker recognition systems use semantic analysis, though
this applies not only to systems of this type. It is used with equal success in cate-
gorisation processes, e.g. of medical systems.
