(Continued from page 5)

urements would result in inaccurate calculations of mean F0 and standard deviations.  Of course, this meant a considerable amount of additional manual work!
Analysis of F-pattern presented the greatest challenge in the investigation.  Firstly, it is imperative to compare linguistically equivalent data.  In both cases, the aircraft's identification call name proved extremely useful for this, since it was included in almost all trans

in each word.  Such a simple solution was not the case with the other aircraft, and instead three formant measurements were made at voice onset, at 50% through the voiced segment, and at voice offset.  The resulting measurements from both pilots were plotted on individual vowel charts, and this way it was possible to compare the "vowel" space of the affected samples, with that of the normal samples from each pilot.
Speech rate was also calculated, by simply timing each utterance, and dividing the number of syllables in the utterance by this time.  In fact both speech and articulation rates were calculated, however the absence of pauses in a number of samples meant that in these cases speech rate was the same as articulation rate.
I am now in the throws of preparing the report for the organisation which requested the analysis.  My final challenge in this forensic investigation is to produce a report which is intelligible to non-speech scientists, while, at the same time provides an accurate account of the acoustic differences between the normal and affected samples of speech.

"...the speech analysis equipment generated a significant number of extraneous pitch measurements."

missions from the pilots to ATC.  This, therefore, provided the data for F-pattern analysis.  However the analysis was restricted to F1 and F2 because the radio transmission frequencies were bandwidth limited to between approximately 250 Hz and 2250 Hz.  Because the tokens of the aircraft ID varied considerably in length, comparing the affected sample with the normal samples was a rather messy task.  With one data set, however, the three words in the aircraft ID had the syllable structure C1V1C2V2, where C2 was either /r/ or /l/.  As a consequence F2 had a falling-rising contour in each of the three words of the call ID.  It therefore seemed appropriate to take formant measurements at three points, based on the highest, the lowest and the highest F2 measurements