THE POWER OE FUNDAMENTAL SPEECH SOUNDS 397 



Our measurements of mean power, on the other hand, were made 

 from power records of conversational speech, with a greater variety 

 of observations and speakers. Stress, therefore, plays an important 

 part here. 



In Table II is given a compact summary of the direct measurements 

 made on the power oscillograms. Thus consider "d" as in "dot." 

 2.9 microwatts was the greatest observed value for any speaker, while 

 the average of all observations (including accented and unaccented 

 utterances) was but 0.08. Only four observations, however, were 

 large enough to be measured. As before, we give the corresponding 

 intensities in microwatts per square centimeter at the transmitter in 

 the next two columns. 



To show the occurrence of stress in the utterance of these sounds in 

 ordinary speech, we give in Fig. 1 the stress frequency -distribution 

 curves ° of several oft-occurring sounds. These curves are derived in 

 the same manner as were the syllabic stress curves in the study of 

 speech power. They exhibit the marked degree in which the conso- 

 nants differ in stress for ordinary speech. For example, among the con- 

 sonant sounds, "t" and "sh" represent extreme types. The former is 

 either slighted or strongly accented with but little intermediate grada- 

 tion while the blunt characteristic of the latter indicates the most 

 nearly uniform distribution of stress into all shades from zero to maxi- 

 mum. Similarly with the three semi-vowels shown, "1" and "m" are 

 extreme types. 



The Vowels 



Some attention was given to vowel power in the other paper where 

 under the heading of "Relative Power of Vowels" (on page 634) were 

 charted what we have classified as normal values of mean power. 

 These were derived from the mean power curves of disconnected mono- 

 syllables. Although they were charted separately for male and female 

 voices, we shall not differentiate between the two in the following. In 

 Tables III and IV are summarized the four sets of data based upon the 

 speech from 16 voices. Here we see the influence of stress by comparing 

 the conversational and normal values. This effect is noteworthy in 

 the case of "o" (ton) "a" (tap) and "i" (tip) which average consider- 

 ably less power in conversational speech than in normal syllables. 

 Another point of interest is the comparison of peak and mean values. 

 For example, in the normal data, the ratio of peak to mean (i.e. the 



* The abscissa represents the relative number of observations (s/s) whose relative 

 power values exceed the magnitude of the ordinate, n, a numeric varying between 

 zero and one. 



