Klaske van Sluis

The acoustic contrast between the Dutch consonants /t/ and /d/ 73 4.2.2 Data collection Recordings are made during regular appointments at the speech language pathol- ogist. Over time, three different microphones were used during consultations; HS5 Samson Aerobic Headset Mini Jack/XLR STAGE 55/CONCERT 77/AKG Version, Shure SM10A-CN headset with a Blue Icicle USB microphone pream- plifier and Samson Qv10e micro-phone. Participants were instructed to pro- nounce a sustained /a/ as long as possible and to read aloud a Dutch text of neutral content ( Tachtig dappere fietsers ’eighty brave cyclists’). In this text, 18 token words occurred (9 starting with /t/, 9 starting with /d/). For two participants, only the first and last two sentences of this text were recorded, since they had difficulty reading the text in its entirety resulting in 11 tokens. 4.2.3 Acoustic analysis Speech recording data were processed in Praat [8]. Voicing characteristics be- fore and after treatment are analyzed on sustained /a/, to check whether voice contrast analysis is possible. All /a/ recordings were cut to a length of 2.0 seconds of the most stable part (as done in van As-Brooks et al. [3]), to en- sure optimized and similar data for all participants. The measured acoustic parameters are: F0 (median/SD), percentage voiced (%V), Maximum Voicing Duration (MVD), and Harmonics-to-Noise Ratio (HNR). Pitch was determined with the standard settings of the To Pitch command using the cross-correlation method with the exception of: pitch floor 40 Hz, ceiling 200 Hz; voicing thresh- old 0.40. The median F0 was determined as the median value of all the voiced frames. %V was determined by counting the number of voiced and unvoiced frames as determined by the Pitch algorithm. MVD was taken as the longest voiced segment after an automatic annotation with the To TextGrid (vuv): 0.2, 0.1 command. HNR was calculated adapted from van As-Brooks et al. [3] as the smoothed maximum found after calculating To Harmonicity (cc) with a minimum pitch of 40Hz, silence threshold 0, and 4.5 periods per window (1 period per window if no HNR was found). For each t/d token, the begin and end of the release burst were marked (at nearest zero-crossing). The length of the release burst was measured. The presence of an F0 value inside a 35ms window before the start of the release was considered evidence of prevoicing. A window of 35ms was chosen to preclude an influence of the preceding word. Vowel duration was calculated from the end of the release burst to the vowel offset. 4.2.4 Statistical analysis All data is collected and processed for statistical analysis in R [9]. A Kolmogorov- Smirnov test showed the data was not normally distributed. Therefore, the Wilcoxon Matched Pairs Signed Rank test (WMPSR) is performed to analyse differences between pre- and post-total laryngectomy voice recordings. Sta-