TABLE IV.
Machine learning classification accuracy (using 5-fold cross-validation) within individual utterances. Data were from exp. 3. Bold entries are significantly above chance performance (50%) at a 0.05 level according to a one-sided binomial test.
| Talker | /ɑtʃɑ/ | /ɑdʒɑ/ | /ɑdɑ/ | /ɑtɑ/ | /ɑfɑ/ | /ɑvɑ/ | Avg. |
|---|---|---|---|---|---|---|---|
| W3 v1 | 54.5 | 65.3 | 58.9 | 49.0 | 53.8 | 54.2 | 55.9 |
| W3 v2 | 58.0 | 68.0 | 65.5 | 62.5 | 58.5 | 59.6 | 62.0 |
| W3 v3 | 64.4 | 66.2 | 58.5 | 66.5 | 64.0 | 56.6 | 62.7 |
| W2 | 57.8 | 49.5 | 73.2 | 62.7 | 52.3 | 55.0 | 58.4 |
| W4 | 60.0 | 61.0 | 54.1 | 71.7 | 57.6 | 58.5 | 60.5 |
| W5 | 59.5 | 61.2 | 56.7 | 64.2 | 52.6 | 53.8 | 58.0 |
| Avg. | 59.0 | 61.9 | 61.1 | 62.8 | 56.4 | 56.3 | 59.6 |