SPEECH POWER AND ITS MEASUREMENT 



651 



decidedly fatiguing to maintain a speech level 20 db higher for more 

 than a few seconds. 



It should be noted that the distributions of the instantaneous 

 power and instantaneous pressure values in respect to their average 

 values are quite different for speech from what they are in a sinusoidal 

 wave. Fig. 1 and Fig. 2 show the extent of the difference for powers 

 and pressures respectively. Clearly, speech as a whple is decidedly 

 more "peaked" than a sinusoidal wave. Values of peak factors for 



■20 



-50 



100 1000 



FREQUENCY -CYCLES PER SECOND 



Fig. 3 — Energy spectrum of connected speech-composite curves for 4 'men and for 



2 women. 



individual speech sounds, i.e. the ratio of peak amplitude to mean 

 effective amplitude, are given in Sacia's paper. ^ 



A paramount characteristic of speech is the distribution of 

 power and pressure in the frequency spectrum. That is so because 

 in general the systems responding to or carrying speech (including 

 the ear) have pronounced frequency characteristics. The same is 

 true of noise sources interfering with speech. Here, as for total 

 powers and pressures, the simplest quantity is the average spectrum 

 for a large number of speech sounds. Crandall and Mackenzie ^ 

 obtained the energy spectrum shown in Fig. 3, based on speech from 

 six voices. They used a condenser microphone whose output was 

 analyzed by a series of resonant circuits covering the range from 75 

 to 5000 p.p.s. Fig. 4 shows a recent determination in which the 

 average pressure is given as a function of frequency. The apparatus 



