396 



DELL SYSTEM TECHNICAL JOURNAL 



corresponding peak powers. Now the approximate peak powers of 

 these vowels for the two speakers were found (see note under Table I) 

 from the power oscillograms used in our study of speech power. Hence 

 from the product we derive the approximate peak power of the con- 

 sonant (or semi-vowel). Direct measurement of peak power from the 

 latter oscillograms was impracticable because of the low sensitivity of 

 the instantaneous power recorder ' and the before-mentioned fact that 

 the power of the consonants and semi-vowels is low relative to that 

 of the vowels. 



Since frequencies of the order of 50 cycles are of negligible importance 

 in speech, the 50-cycle puff has been separated from the other compon- 

 ents in the case of the unvoiced stop consonants. This is justified by 

 the fact that the utterances of such a sound by two speakers may seem 

 exactly alike to the careful listener, whereas a large puff may be present 

 in one case and none in the other. 



The values thus far considered represent "normal" values in speech 

 — not accented and yet not slighted. 



TABLE II 



Conversational Values of Mean Power in Microwatts for 16 Speakers 



(A) Consonants 



(B) Semi- Vowels 



' In recording the power, separate vibrators had been used for instantaneous and 

 mean powers. 



