FIG. 4.

(Color online) Percent errors (mean and standard deviation from experiment 1) for each phonetic category for intact and vocoded SiB (A) and intact and vocoded SiQuiet (B). The labels “Voicing,” “Manner,” and “Place” correspond to when the consonant reported differed from the consonant presented only in voicing, MOA, or POA, respectively. “All V,” “All M,” and “All P” correspond to when the consonant reported differed from the consonant presented in at least voicing, MOA, or POA, respectively [e.g., “All V” includes the following types of errors: (i) voicing only, (ii) voicing and MOA simultaneously, (iii) voicing and POA simultaneously, and (iv) voicing, MOA, and POA simultaneously]. The expected distribution of errors under the null hypothesis of random confusions was generated separately for (A) and (B) and with 1000 realizations each. Each realization of each null distribution was produced by generating a Bernoulli trial with “success” probability = 60% for (A) or 90% for (B), followed by uniform-random selection of a different consonant from what was presented if the trial outcome was “failure.”