Table 5 Results (TER%) of Transformer-based E2E CS ASR systems with different target acoustic modeling units
From: Pronunciation augmentation for Mandarin-English code-switching speech recognition
Modeling units | Mandarin | Code-switching | Chilish |
|---|---|---|---|
Baseline (character-letter) | 5.2 | 11.6 | 15.9 |
Character-BPE | 5.3 | 11.1 | 15.3 |
Character-PASM (CMU+MLex) | 5.2 | 10.7 | 15.3 |
Character-PASM (E2M-po2m) | 5.1 | 10.5 | 15.1 |
Character-PASM (Source Lexicon+Lexlearn (2.41)) | 5.2 | 10.1 | 14.8 |