614 BELL SYSTEM TECHNICAL JOURNAL 



not the sizes of these curves which are significant. The order in which 

 these curves are arranged is based upon the vowel triangle, and on 

 Table I\'. To return to the general discussion, we find that the 

 fundamental voice frequencies do not have large efTective amplitudes; 

 it is interesting to note that these can be largely eliminated without 

 impairing the distinctive quality of a vowel sound. The "scattered 

 low frequencies" of the table (Sounds I to \TI) exhibit appreciable 

 amplitudes in the diagram. The "Scattered High Frequencies" of 

 sounds I-YII previously noted exhibit small amplitude in the diagram. 

 These are perhaps not essential to these speech sounds, but we should 

 expect to find them in well trained singing voices. They are to a 

 certain extent (particularly for the male voices), paralleled by the 

 high-frequency regions of resonance for these sounds given in Paget's 

 diagram, to which reference was made in Section I. Paget, it must 

 be noted, is convinced that these high frequency regions of resonance 

 are characteristic of the sounds of Groups I-VI. 



The sound a (No. VI) is as it were the center of gravity of the vowel 

 diagram and occupies the key position in the phonetics of most lan- 

 guages. The broad feature of the diagram is of course the progressive 

 rise in frequency and gradual narrowing in range of the characteristic 

 region of resonance, till the sound a is reached, succeeded by a splitting 

 up into two regions of resonance which recede from one another as we 

 follow the diagram downwards from a to the end. The e.xact location 

 of sound X (er) is somewhat indeterminate, but it is evident that it 

 belongs in the series of doubly resonant vowels. It is interesting to 

 note that the distribution of the components of ar (refer either to 

 Table IV or Fig. 13) is similar to the distributions gi\en by Miller and 

 by Paget for a form of the vowel a having "double" resonance; it is 

 therefore as well located as any vowel in the series. 



The characteristics of the r sound (whether considered as \owcl 

 or consonant) offer an interesting study, and in considering them 

 we have an illustration of the practical value of records of the 

 type shown. The problem of pronouncing a pure r sound is diflicult; 

 r is probably as variable in quality as any sound in the language, and 

 it differs more than any other sound from one language to another. 

 The precise location of its characteristic frequencies is thus a rather 

 diflicult matter. The records of ar and er disclose a noticeable tendency 

 in speaking to make these sounds into diphthongs, the earlier portion 

 of the record being nearly a pure a or (short) e while the latter portion 

 of the record increasingly displays r characteristic. One sj^eaker (MA) 

 succeeded in making records for these two sounds which have nearly 

 the same character throughout (Plates 49, 73), but for the other seven 



