246 



VOICE COMMUNICATION, II 



db less than that pro\aded by the narrower 

 band. Now, suppose that the signal-to- 

 noise level at which both systems give 10 

 percent articulation can be improved 5 db. 

 This improves the narrow system from 10 

 to 20 percent, but improves the wider system 

 from 10 to 50 percent articulation. 



This family of curves is constructed with 

 1500 cps as the geometric center frequency 

 of the band, but to a reasonable approxima- 

 tion the same contours can be used for other 

 center frequencies by shifting the entire fam- 

 ily horizontally one way or the other on 

 the semi-log plot. 



IV. AmpUivde Selectivity 



Amplitude, or non-linear, distortion occurs 

 whenever a signal is passed through a cir- 

 cuit which transmits some amplitudes of 

 the wave and not others. The most fre- 

 quent type of amplitude selectivity is due 

 to circuits which transmit low-amplitude 

 parts of the wave more efficiently than high- 

 amplitude parts. Reduced to its simplest 

 form, this type of distortion is called peak 

 clipping — the peak voltage swings are 

 clipped symmetrically off the top and bot- 

 tom of the speech wave. 



Although peak clipping degrades the qual- 

 ity and naturalness of speech, articulation 

 tests indicate that a surprisingly large frac- 

 tion of the speech wave can be eliminated 

 before intelligibility is affected (10, 13, 19, 

 28). The resistance to the effects of peak 

 clipping is sho^vn graphically in Fig. 3. 

 More than 95 percent of monosyllabic words 

 are recorded correctly even after 24 db of 

 peak clipping, when all that remains of the 

 speech wave is the re nearest the center 

 axis. About 70 percent of the words are 

 understood after infinite peak clipping which 

 reduces speech to a succession of rectangular 

 waves (21). 



If the system passes the high-amplitude 

 parts of the wave and rejects the low-ampli- 

 tude parts, it is a case of center clipping. 

 The center of the wave nearest the time 

 axis is clipped out and thrown away. With 



it goes the intelligibility of speech, as is evi- 

 dent from Fig. 3. Clipping out only one- 

 quarter of the speech wave (2.5 db center 

 clipping) reduces word articulation to 30 

 percent. This type of distortion results in 

 sounds that resemble static more than 

 speech. The disastrous effect of center clip- 

 ping is due principally to the elimination of 

 the faint consonant sounds which provide 

 most of the intelligibility. Peak clipping, on 

 the other hand, tends to emphasize the con- 

 sonants relative to the vowels. 



Some types of amplitude selectivity are 

 intolerable, while other tj^Des can be ac- 

 cepted and even used to practical advan- 



CLIPPING IN DECIBELS 



Fig. 3. The effects upon articulation scores of 

 clipping the peaks versus clipping the center out 

 of the speech wave 



tage. Peak clipping circuits have actually 

 been recommended in some situations — e.g., 

 in amplitude-modulated radio transmitters 

 (15, 18, 23, 26), radio receivers (19), and 

 hearing aids (7). 



V. Evaluation 



We have touched only the high spots in 

 this outline of the effects of masking and 

 distortion upon voice communications. 

 Many other masking signals and types of 

 distortion have been studied and many re- 

 main for future study. This work should, 

 and probably will be, continued. Every 

 new communication device brings novel 

 masking sounds, new possibiUties for dis- 



