References:
1.
Фу К. Структурные методы в распознавании образов. – М.: Мир, 1977.
2.
Фант Г. Акустическая теория речеобразования. – М.: Наука, 1964.
SIGNS OF SIGNALS OF THE SPEECH SIGNAL
IN THE SYSTEMS OF RECOGNITION OF SPEECH
G.Qayumova (TUIT, Senior teacher),
D.Shamuratov (TUIT, student)
In recent years, there has been a sharp increase in interest in the development and use of
algorithms for processing speech signals. The use of modern computers with colossal computing
powers at the present time in the presence of significant operational memory and flexible external
devices makes it possible to solve the task of processing speech signals in almost real time, which
led to the widespread use of computer systems that to some extent use speech technologies.
It is known that the speech signal itself changes with time rapidly enough. This is due to
the peculiarities of the formation of a speech signal by filtering the signals of excitation of air
pressure pulses, air shocks arriving from the vocal cords as they vibrate, i.e. through open vocal
cords from the lungs during exhalation through the resonance system (articulatory organs - larynx,
tongue, mouth cavity and Nose). But, the properties of the speech-forming tract vary slowly
because of its inertia. Therefore, in speech recognition systems, it is necessary to receive and store,
in a digital form, slowly changing parameters of the voice path and the source - the pitch
frequencies, the formant frequencies, which determine the character of the speech signal itself. In
this connection, this work considers the issues of determining the parameters of the speech-
forming path-formant and various related characteristics, which we will call informative
parameters.
By informative parameters and their changes in time, you can restore the speech wave or
recognize its meaning. These parameters vary little in time and their number is much smaller than
the digital samples of the signal itself. Therefore, you can allocate a smaller amount of memory to
a statement that is parsed or generated. Consequently, less time will require machine processing
when recognizing a speech signal. At the same time, it is possible to provide a system for
recognizing or synthesizing speech with less machine resources, and thereby greatly reduce its
cost.
The report will consider methods of primary digital speech processing. These methods
underlie modern systems of automatic speech recognition and synthesis and are associated with
obtaining current autocorrelation of the signal, energy spectrum, linear prediction parameters,
homomorphic processing, and clipped speech.
257
Do'stlaringiz bilan baham: |