EFFECTS OF PHASE DISTORTION 555 



it were made up of two sets of networks, the two sets being connected in 

 series. The first set consists of a number of networks in parallel, each 

 network passing a frequency range A/ and having a phase characteristic 

 of the form Bxw, where J5i is the slope of the straight line approximating 

 the curved phase characteristic in the range A/. The second set con- 

 sists of a number of corresponding networks having phase characteris- 

 tics of the form Bq, where Bq for a network passing the range A/, is the 

 constant term in the equation of the straight line approximating the 

 curved phase characteristic in this range. The phase distortion thus 

 consists of two operations in sequence, the first introducing definite 

 delays in various portions of the original wave, the second introducing 

 constant phase shifts in the component waves of each delayed portion. 



A definite contribution to the recognizability of a speech sound 

 may be associated with each frequency range A/ in the undistorted 

 state. At the output terminals of the first set of networks the various 

 portions corresponding to the ranges A/ do not combine to form an 

 exact copy of the original wave, because of the different delays that 

 have been introduced. It is supposed that their normal contributions 

 to recognizability are decreased by a factor depending upon the delay 

 distortion and the duration time of the speech sound. This factor is 

 referred to here as the " time factor " and it would be expected to operate 

 even though the second set of networks were non-existent. 



Since the constant phase shifts of the second set of networks are not 

 all multiples of It, additional distortion will be introduced by this set 

 of networks. To take account of this another factor called the "inter- 

 cept factor" is introduced. As will be seen later this factor seems to 

 be negligible for the case of speech waves due in part, no doubt, to the 

 sustained character of the waves and to the mechanism of hearing as 

 previously described. 



In addition to the above, when we deal with a succession of speech 

 sounds, as in connected speech, a third factor might be expected to 

 operate because the delayed frequency ranges of one sound may overlap 

 the least delayed ranges of a succeeding sound and interfere with its 

 perception in the manner of an extraneous noise. As will also be seen 

 later on this factor appears to be negligible for the type of characteristic 

 shown in Fig. 2 because the so-called noise and the signal with which 

 it interferes do not have components in a common frequency range. 

 When this is true, noise in general interferes much less than when the 

 signal and the noise have components in a common range. 



Phase Distortion and Quality 

 Quite aside from the recognizability viewpoint, when speech from a 

 system having phase distortion is compared with that from a system 



