ENERGY DISTRIBUTION IN SPEECH 119 



entire electrical response of tl.c system; the other side suppresses 

 all save a band of frequencies, the center of this hand being shifted 

 by resetting the condenser and inductometer. 



Having chosen for analysis a piece of connected discourse, the 

 speaker utters the successive syllables separately but as nearly as 

 may be with the same inflection and volume as if the syllables were 

 continuously spoken. Two observers record the readings of microam- 

 meters in the couple circuits of the thermocouples. One of these 

 instruments gives a deflection corresponding to the total energy of 

 the syllable uttered; the deflection of the other instrument corre- 

 sponds to the energy of the syllable lying within the limits of trans- 

 mission of the tuned circuit. 



Preliminary experiments were carried out to determine the relation 

 between momentary deflection read on the microammeter, and the 

 current momentarily flowing in the thermocouple heater. Currents of 

 different values were caused to flow for intervals of time varying 

 from 0.2 second to 1.2 seconds, and the deflections were found nearly 

 proportional to the product of current squared and time interval; 

 this proportionality was most nearly exact when the current was 

 weak and the time intervals short. For all cases likely to be dupli- 

 cated in the speech analysis work the error might be taken as about 

 5 per cent, a quantity small in comparison with the inevitable un- 

 certainties due to other causes. 



Quite low damping is attained in the resonant circuit. The values 

 of inductance used ranged from 0.20 to 0.66 henry and the total re- 

 sistance of the circuit— transformer winding, inductometer coil, 

 thermocouple heater— is of the order of 100 ohms. The damping 

 thus ranges from 75 to 250. 



The circuit is calibrated in the following manner: 



A switch is so introduced that it is possible to include in series with 

 the thermocouple the resonant circuit, or replace it by a non-in- 

 ductive resistance whose value is approximately that of the A. C. 

 resistance of the inductometer winding. With the tuned circuit 

 excluded, an alternating current of suitable magnitude is caused to 

 flow in the thermocouple heater; the tuned circuit is then substituted 

 and the new value of the current observed, the input voltage remaining 

 constant. The ratio of current squared "tuned circuit in" to current 

 squared "tuned circuit out" is plotted against frequency, yielding a 

 curve for energy transmission. 



Twenty-three bands in all were considered adequate for the analysis 

 of energy distribution in speech; the centers of these were at 75, 100, 

 200, 300 cycles, 400 to 3,200 cycles by steps of 200; 3,500, 4,000, 4,500, 



