Expert profile: Michael Fateev

---

Description: Please see for more information in My Profile.

      New basis for recognition, synthesis and compression of a speech

                                   signal


                             Michael K. Fateev

   Our experience in the field of recognition and synthesis of a speech
   signal is more than 30 years in total. We have checked up all known
   methods of the primary description of a speech signal such as Fourier
   Transforming, band filtering, LPC analysis (Linear Predictive Coding),
   MFCC (Mel Frequency Cepstral Coefficients) and others. All these kinds
   of description depend on such individual factors as timbre of voice,
   position of the phoneme, intonation etc. Last year we managed to
   receive the invariant description of a speech signal, which
   practically does not depend on these variances and corresponds to the
   description which the human ear uses. As far as we know, no one has
   received such results before.

   To show opportunities of this description we have developed the
   software demo-version ensuring recognition of six Russian vowels.
   These best-in-the-world results in recognition of separate phonemes
   demonstrates SpeechViewer of IBM (http://www.IBM.com).

   To objectively estimate parameters SpeechViewer it is necessary to
   consult with those who use this program (school for deafs). If you are
   an individual with good hearing, after long and complex procedure of
   selecting your voice samples, the software can distinguish your
   phonemes quite enough. However, if in you try again to test the same
   software some time later, the accuracy of recognition may have sharply
   decreased. When a deaf person works with the software (children in
   particular), it often perceives the bad sound as good and on the
   contrary.

   Our demo-version (http://www.sounddescription.webjump.com) is a game,
   which is as though a continuation of a set of SpeechViewer, but its
   characteristics are unattainable for the competitor. This software
   distinguishes phonemes without special samples for your voice. The
   accuracy of our product requires only a single version of phonemes to
   be stored in order to assess arbitrary reported speech of any
   announcer including children and whispered speech.

   Our product not only distinguishes phonemes, as is done by
   competitors, but determines quality of your pronunciation. For
   increased passion and for demonstration of opportunities the game is
   made in such a manner that it is very difficulty to receive high
   scores even for good announcers.

   Our Email: fateevs@usa.net.