Recording procedures Measurements and Plots

26 while avoiding nasal consonants, for each of the wordlists I freely chose other words with consonants from a variety of places and manners of articulation. Aside from the consonant issues, the criteria for choosing words differed somewhat depending on the wordlist, since each list was designed with the intention of analyzing vowels in a specific position or word type when surrounded by vowels of varying qualities. Therefore, more specific criteria concerning the number of words used, vowel length, parts of speech and noun classes, will be addressed in each individual section where measurements from different wordlists are discussed.

1.6.3 Recording procedures

Recordings were made using an Audio Technica AT899 condenser microphone and the MicPort Pro preamp. The microphone was mounted onto a headset and placed approximately an inch to the side of the speaker’s mouth. I primarily used WavePad Sound Editor 4.42 to record directly to my laptop computer at a sample rate of 48,000 Hz, 16 bit, mono. The Praat sound recorder was used for a small number of the sound files during the beginning of the recording process with Speaker B. Recording locations varied considerably. All recordings of Speakers A, B and C were made in various rooms in an office building in Musoma, Tanzania. A sound-proof room was not available, but efforts were always made to find as quiet of a location as possible. All recordings of Speaker D were made in a quiet location indoors in the Ikoma village of Morotonga, just outside of Mugumu, Tanzania.

1.6.4 Measurements and Plots

I used SIL’s Speech Analyzer software for all vowel formant and bandwidth measurements. For each vowel, I measured the first and second formants F1 and F2 27 using the LPC 18 display. For most vowels I also measured the first formant bandwidth B1. Measurements were taken using a single wave cycle near the center of the steady state portion of the vowel. I made an effort to choose a cycle with formant values which were most representative of the steady portion of the vowel as a whole. When measurements varied significantly from one cycle to the next, I often used Speech Analyzer’s formant tracks to get an idea of which LPC measurements were most likely to be accurate. The formant tracks are calculated differently than the LPC, so I opted for cycles in which the LPC and formant track measurements were in close agreement. I used Microsoft Excel workbooks to record and organize all measurements. All vowel plots were made using the FPlot software. 19 I used the website http:www.physics.csbsju.edustatst-test.html for performing unpaired t-tests.

1.7 Overview