SPEECH POWER AND F.NEKGY 639 



pro<U-tormiiu'(l points by tapping a key which momentarily displareil 

 the timing wave. By the aid of these punctuations we were enabled 

 to identify the words and syllables on the records after development. 

 The areas for computing average power W'ere measured from the mean 

 power curve, while the instantaneous power curve was measured only 

 for its peak values. 



Although chosen at random, the speakers used in these tests repre- 

 sent all sections of the United States. Their types range from soprano 

 to bass-baritone, neither extreme type — high soprano and bass — being 

 available; but this assortment is sufficiently representative for our 

 purpose. Extraneous disturbances were to a large extent eliminated 

 by the sound-proofing on the walls and ceiling. Lest the novelty of 

 this situation be a distraction to the speaker, he was allowed to prac- 

 tice and become accustomed to the new condition. 



Conclusion 



One advantage in having speech data available in terms of its power 

 rather than its amplitude is the fact that in most instruments used for 

 making quantitative speech measurements, the force which operates 

 the meter is proportional to the square of the wave amplitude. Com- 

 mon examples of such instruments are the dynamometer and the 

 ordinary a.c. meters. 



To summarize, the power is classified into: 



1. Instantaneous power, P,. 



2. Mean power, Pm- 



3. Long average power. 



4. Short average power. 



Stress calibrations are here derived from the maximum values of Pi 

 and Pm (Pi and Pm, respectively) in each syllable, while the use of the 

 total energy of the syllable for calibrating emphasis also shows possi- 

 bilities. The peak factor is the square root of Pi! Pm and is a useful 

 index of the wave form. 



The measuring apparatus — excluding the rectifier and oscillograph — 

 is essentially a good quality speech-transmitting system. In view of 

 the fact that good quality systems are now used commercially as well 

 as in the laboratory the data naturally fall into two classes: 



(1) Measurements which characterize the speech solely from the 

 standpointof the transmitting apparatus; 



