244 



VOICE COMMUNICATION, II 



II. The Masking of Speech 



The effect of an interfering noise is to 

 raise the thresholds of detectabihty and in- 

 telligibility, and so to decrease the range of 

 intensities available for vocal communica- 

 tion. This shift in the threshold is the meas- 

 ure of masking. Military communications 

 are chronically afflicted with noise, and an 

 estimate of how much noise can be tolerated 

 in any given situation is a highly practical 

 undertaking. 



Pure tones are relatively ineffective in 

 masking speech. If they are intense enough, 

 low-frequency tones can mask the entire 

 range of speech frequencies, but high-fre- 

 quency tones mask only a restricted range 

 of frequencies. The listener is able to deci- 

 pher the speech on the basis of the unmasked 



GUT-OFF FREQUENCY IN CYCLES PER SECOND 



Fig. 1. The relation between articulation score 

 and the cut-off frequency of high-pass and low- 

 pass filters. Composite for men's and women's 

 voices. (After French and Steinberg, 12) 



portion of the speech spectrum (27). Com- 

 plex tones of low frequency are more dis- 

 rupting, but the greatest danger comes from 

 continuous spectrum noise. 



The simplest type of noise to specify for 

 experimental purposes is one which has a 

 uniform level per cycle over the range of 

 speech frequencies (100-8000 cps). This 

 type of noise is sometimes called white noise. 

 When the sound pressure level of this noise 

 is 40 db, the threshold is shifted about 12 

 db. For levels above 40 db, this noise pro- 

 duces a one db shift in the threshold for every 

 one db increase in the intensity of the noise 

 (16). This relation is sometimes expressed 

 by saying that the signal-to-noise ratio nec- 



essary for communication is constant over 

 a wide range of intensities. Only at very 

 low and very high noise levels must the 

 signal-to-noise ratio be increased. Expe- 

 rience has shown that for most continuous 

 spectrum noises encountered in military sit- 

 uations the signal-to-noise ratio should ex- 

 ceed +6 db for satisfactory communications, 

 although the presence of speech is usually 

 detectable at ratios as low as — 18 db. 



Aspects of the noise which are known to 

 influence its masking effectiveness include 

 its spectrum (22), its temporal continuity 

 (22), the phase relations between the ears 

 (20), and the meaning of the noise to the 

 listener. Further research is needed to as- 

 sess the importance of these variables. 



III. Frequency Selectivity 



Voice communication equipments were 

 markedly improved during World War II, 

 especially as regards the reduction of fre- 

 quency distortion. A great deal of fre- 

 quency distortion can be tolerated under 

 otherwise ideal conditions, but a system 

 which passes a narrow range of frequencies 

 quickly shows its inadequacy when tested 

 under the stress of noise. 



In laboratory tests of frequency selectivity 

 it is desirable to use filters which pass uni- 

 formly a given range of frequencies and 

 greatly attenuate all components of the sig- 

 nal which lie outside this range. Although 

 such systems are rarely encountered in prac- 

 tice, they have the advantage of enabling us 

 to estimate the relative importance of the 

 various frequencies in the spectrum of human 

 speech. 



The results of articulation tests conducted 

 in the quiet with nonsense syllables distorted 

 by high-pass and low-pass filters are sum- 

 marized in Fig. 1 (12). If a high-pass filter 

 is used to select only components above 

 6000 cps, no syllables will be correctly re- 

 corded by the listeners. When only fre- 

 quencies above 3000 cps are passed, 30 per- 

 cent are discriminated. Frequencies of 1000 

 cps and above give 90 percent syllable artic- 



