PHYSICAL CHARACTERISTICS OF SPEECH AND MUSIC 365 



These values confirm earlier results obtained by oscillographs and 

 give a much more detailed picture of the variation of the peak values 

 as the speech proceeds. About 2 per cent of the time the peak power 

 in l/8th second intervals exceeds the average power level by 20 db; 

 that is. it is more than 100 times greater. It is seen that a system 

 designed to transmit conversational speech of the best quality should 

 be capable of handling at least 1000 microwatts instead of 10 micro- 

 watts. It is also seen that the most frequently occurring peak is at 

 about 10 times the average speech power. For 21 per cent of the time 

 the peaks are below the average level. A large number of the l/8th 

 second intervals in this class are silent. 



To find how the speech powers are distributed throughout the 

 pitch range similar measurements were made introducing successively 

 each one of the 14 band filters as indicated in Fig. 12. These bands 



-2 -1 



PITCH-P 

 OCTAVES FROM 1 KILOCYCLE 



Fig. 15 — Distribution function for conversational speech. 



SdP. 

 Pi 



were arranged so as to cover about 1/2 octave pitch range except at 

 the two lower octaves where they cover a complete octave. From 

 the measurements on the average speech power in each band the curves 

 in Fig. 15 were constructed. They give the results for average con- 

 versational speech for both men's and women's voices. The ordinates 

 are such that the fraction of the total power F which is carried by any 

 pitch interval between Px and P2 is given by 



Jp, 



F = I 10^- dP. 

 Ipi 



(7) 



