Real and artificial speech signals for objective quality evaluation of speech coding systems
Kenzo Itoh, Nobuhiko Kitawaki · Electronics and Communications in Japan (Part III Fundamental Electronic Science) · 1989
Abstract This paper discusses the test signals to be employed in the objective evaluation of the speech coding system. the dependency of the objective evaluation measure on the speaker is examined. the feature parameters of the speech, which are largely responsible for the dependency, are discussed. the number of speakers and the length of the speech sample required in the stable objective evaluation are examined. As the result, it is concluded that one should be careful about the dependency on the speaker when the real speech signal is used as the test signal. It is desirable to use a speech sample of longer than 4 to 5 s uttered by some 10 speakers. Then the test signal other than the real speech is discussed. Artificial speech, which approximates the distribution of the speech signal on the frequency domain, is discussed. A new artificial speech (ASVQ) is proposed based on the vector quantization and the speech synthesis technique. to verify the usefulness of ASVQ, examinations are made for a typical speed coding system using distortion measures in the time and frequency domains. As a result, it is seen that ASVQ represents the features of the speech more efficiently than other artificial speeches which have been proposed for various purposes. It is shown that by employing ASVQ as the test signal, an objective evaluation very close to the value by the real speech can be obtained.