Table 1.
Classification accuracy of spikes from both hardware (HW) and software (SW) cochlea models using original and reconstructed audios for the training set.
| F/M | Test | |||||
|---|---|---|---|---|---|---|
| Original | HW | SW 1 trial | SW 10 trials | SW 100 trials | ||
| Train | Original | 99.8/100 | 28.2/ 33.1 | 23.8/46.5 | 29.2/54.9 | 31.1/43.3 |
| HW | 92.7/85.5 | 88.1/79.3 | 78.5/73.7 | 84.9/79.7 | 84.5/81.5 | |
| SW 1 trial | 90.3/82.3 | 71.9/61.3 | 79.3/80.3 | 89.8/86.3 | 87.9/84.0 | |
| SW 10 trials | 89.3/77.9 | 71.5/63.1 | 87.2/84.1 | 91.2/89.4 | 90.1/87.9 | |
| SW 100 trials | 93.0/77.7 | 71.8/65.0 | 89.3/83.8 | 90.9/89.9 | 90.9/88.6 | |
Results are shown for female (F) and male (M) speakers. The entries in bold show the results when the reconstructed audio in both the training and test sets come from the same cochlea model.