606 BELL SYSTEM TECHNICAL JOURNAL 



In producing this system of records we believe that we have covered 

 the speech sounds as fully as we are justified in doing with the present 

 recording ap[)aratus. In the case of each vowel the combined data 

 from the eight records constitute a sufficient basis for the most thorough 

 harmonic analyses that can be made and they should yield accurate 

 results for the characteristic vowel frequencies. In analysing these 

 records small corrections are of course necessary on account of the 

 slightly imperfect frequency characteristics of the apparatus, but 

 these corrections can be taken without difficulty from the calibration 

 curves. 



The amplitude scale in these records is arbitrary in each case. This 

 is for the reason that, owing to the widely different conditions of voice 

 control among the different speakers, the recording apparatus had 

 to be adjusted to different levels of sensitiveness for each record in 

 order to obtain the requisite maximum oscillation of from 1 to 2 

 centimeters. No attempt has been made to compare the absolute 

 amplitudes from one record to another on account of these intensity 

 variations. The emphasis has been placed rather on obtaining in each 

 record a good well-defined wave which could be enlarged if necessary. 



Notwithstanding the fact that for frequencies above 5,000 cycles 

 the apparatus was not nearly as good as for frequencies within the 

 calibration range from 75 to 5,000 cycles, the records obtained of some 

 of the consonant sounds are of considerable practical \'alue. It is 

 felt however, that the present apparatus has been used nearly to the 

 limit of its possibilities and that devices other than the usual oscillo- 

 graph vibrator offer more promise in any further investigation of the 

 consonant sounds. It is planned later to issue a more complete set of 

 these records as a supplement to the present paper in order to make 

 the collection a\ailahk- lo ihosi- t'spccialK- interested. 



IV 



Statistk Ai. -Sudv and Harmo.nic Analysis of thk 

 \'<)\vi;i, SoiTNDs 



A detailed inspection of the records takiii, and particularh- of the 

 records of the vowel groups shows thai iniich labor would be required 

 to anaK'ze these records throughout their length, according to the 

 usual methods of harmonic analysis. In nearl%- e^•ery case it wduhl 

 be imp^jssibie to obtain the mean energy distribution in a gi\i'n 

 record, allowing for variations from cycle tcj cycle of the fundamental, 



