Table 8 Relative improvement depending on the content and style layers in cases of ASR model-based evaluation using 10% UME-ERJ subset
From: Accent modification for speech recognition of non-native speakers using neural style transfer
Style layers | Content layers | RI(CER) |
|---|---|---|
1–10 | 6–12 | 32% |
1–8 | 8–12 | 29.6% |
1–10 | 10–12 | 30.1% |
1–5 | 5–12 | 26.7% |
1–4 | 4–12 | 15.6% |