Harmonic Amplitude Spectrums Research Articles

Pitch Synchronous Spectral Representations of Voiced Speech

A number of reduced-information rate spectral representations of voiced speech, analyzed pitch-synchronously, have been examined by digital simulation. The experiment consists of spectral analysis of each period of the voiced portions of connected speech, averaging the spectral information in several ways, resynthesizing the speech from the averaged information, and finally evaluating the result with listening tests. The first method transmits only every other component of the harmonic amplitude spectrum and employs various interpolation functions to recover the absent harmonics. Next, formants are located in two different ways and resynthesis carried out with various numbers and bandwiths of formants. Finally, time averaging is introduced by representing a formant track over a speech event by a polynomial function of time which is the best mean square fit to the track. The resynthesis is carried out varying the number of formants and the degree of the fitting polynomials. The resynthesized samples are evaluated in the listening tests as to their quality and intelligibility. These factors are compared with reduction of information.