A.S. Kolokolov, V.M. Krol, A.Yu. Mestcherakov, I.A. Lubinski, V.P. Yachno

The method for research of informative phonetic cues of the speech signal, combining the analysis and synthesis procedures is offered. It allows to check up influence of separate spectrum fragments of a speech signal on its perception. The method is based on the editing of the dynamic spectrogram of the speech signal, which concludes with removing the separate spectrum fragment and the subsequent restoration of the edited signal in time domain. The conclusion about importance of a concrete fragment of a spectrum in perception of a speech signal becomes on the basis of listening of the initial and restored signals. Spectrum editing is realised in the form of consecutive procedure on which each step the various rectangular fragments of a spectrum allocated for planes "frequency - time" can be suppressed. Operation of removal of fragments of the dynamic spectrogram can be added by insert possibility in the set area of the spectrogram of a fragment from the same or other spectrogram. However it is obvious that this operation is limited only to an insert of noise fragments. It is caused by that at an insert of quasiharmonious fragments in quasiharmonious segments of a signal discrepancy of frequencies of their basic tones leads to specific distortions. They are shown in occurrence beats with the frequency equal to a difference of the basic tones, listened in the restored signal. Their presence increases roughness of sounding of a signal and can complicate listening interpretation. The proposed method can be easy realized by means of modern computer engineering on the basis of fast Fourier transform and represents convenient and flexible tool for speech signal research.