Predictive models of VCV identification and phonetic feature reception based on neural coding of ENV and TFS
| Percent Correct | Voicing | Manner | Place | Nasality | |
|---|---|---|---|---|---|
| Positive SNRs | |||||
| ENV | 0.87 (0.0032) | 1.19 (0.0046) | 1.21 (0.0018) | 1.07 (0.0020) | 1.18 (0.0042) |
| TFS | 0.66 (0.0317) | 1.14 (0.0120) | 0.77 (0.0493) | 0.85 (0.0194) | 1.81 (0.0003) |
| E×T | −0.61 (0.1634) | −1.02 (0.1100) | −0.63 (0.2614) | −0.70 (0.1691) | −1.73 (0.0091) |
| Adj R2 | 0.68 | 0.65 | 0.75 | 0.73 | 0.73 |
| p value | (0.0003) | (0.0006) | (<0.0001) | (0.0001) | (0.0001) |
| Negative SNRs | |||||
| ENV | 0.76 (<0.0001) | 0.30 (0.0028) | NF | 0.40 (0.0042) | 0.40 (0.0560) |
| TFS | 0.33 (0.0011) | 0.19 (0.0031) | NF | 0.11 (0.1705) | 0.82 (<0.0001) |
| E×T | 2.65 (0.0004) | 1.48 (0.0017) | NF | 3.84 (<0.0001) | 3.24 (0.0023) |
| Adj R2 | 0.95 | 0.92 | NF | 0.95 | 0.94 |
| p value | (<0.0001) | (<0.0001) | (<0.0001) | (<0.0001) |
Coefficients from the regression models (i.e., b1, b2, and b3 in Eq. 3, corresponding to ρENV, ρTFS, and their interaction, E×T) are shown with their p values in parentheses, along with the overall model goodness of fit (Adj R2) for overall percentage correct (Fig. 3) and the reception of individual phonetic features (voicing, manner, place, and nasality; Fig. 6). Statistically significant coefficients (p < 0.01) are underlined and bold; marginally significant coefficients (0.01 ≤ p < 0.05) are underlined. Top and bottom sections represent positive (Q, 10, 5, 0 dB) and negative (−5, −10, −15, −20 dB) SNRs, respectively. NF, Not fit (e.g., reception of manner was too poor to fit for negative SNRs).