For each one-feature-different pair of French obstruents (e.g., /p/~/b/), a series of trisyllabic CVCVCV nonwords was created in the following way: For each C position in the CVCVCV word, two nonwords were created using a given filler vowel (V positions were filled with either /a/, /i/, or /u/), such that one of the nonwords contained one segment of the pair and the other nonword contained the other segment. The other two C positions were filled by one of each of the segments. For a segment pair /p/~/b/, for example, a nonword pair was created for each C position (initial, medial, final). An “initial” pair would look like /papaba/~/bapaba/ for the filler vowel /a/, where positions C\(_2\) and C\(_3\) contain one of the segments, but do not vary across the pair.
The Say program developed by Apple, Inc. was used to generate the stimuli. A synthetic process was chosen over natural recordings both for logistical reasons (a total of 432 individual words needed to be recorded by three speakers), and because many of the nonwords were very tongue-twister-like. Given that precision in the order of production of the obstruents was crucial, we opted for the less natural, but more accurate, synthetic solution. The stimuli were generated for each of three synthetic French voices: Sébastien, Thomas, and Véronique.
The duration of the final words was on average XXX ms.