On the informativeness of the extreme octave bands of the speech frequency range

Abstract

The paper presents a review of the topical issues of acoustic speech information protection in terms of the influence of the speech signal’s temporal envelope on the processes of recognition of an intercepted speech under the conditions of existing information and technical confrontation. It is shown that if an intruder who intercepted acoustic speech signals that are limited by the frequency range of the 1st octave which contribution to the indicator of verbal intelligibility according to the current methodological approach of assessing the security of speech information is extremely small, is able to use an automatic speech recognition systems, it becomes possible to isolate the temporal envelope of the original speech signal, to segment it and increase the efficiency of subsequent recognition procedures. By making calculations and experiments the authors have found out that there is a high degree of correlation between the wave formats of the signals in the isolated 1st octave and the full speech signal, as well as between the mixture of the 1st and 7th octaves and the full speech signal. As a result of the analysis several speech parameters that are not taken into account in the current methodological approach are determined. Such parameters are the wave envelope and the first harmonic of the fundamental frequency. According to the results of the experiment those parameters in addition to formants have a significant impact on speech intelligibility. Also is demonstrated the influence of the speech signal waveform (temporal envelope) on the speech signal segmentation procedures in case of a continuous speech stream.

Keywords

• “speech-like” noise, intelligibility of speech, speech signal, information protection, acoustic noise generators.