Table 4 Performance of ASR systems and posterior probability confidence measure
From: Developing a unit selection voice given audio without corresponding text
Training data for ASR | Test data to ASR | %PER | %sub | %ins | %del | %FA | %TR | %FR | %TA |
|---|---|---|---|---|---|---|---|---|---|
Olive | Olive | 1.02 | 0.20 | 0.66 | 0.16 | 0.74 | 0.12 | 0.54 | 98.60 |
LibriSpeech | Olive | 1.57 | 0.58 | 0.69 | 0.30 | 1.00 | 0.27 | 4.94 | 93.79 |
Lecture | Lecture | 5.19 | 0.96 | 3.52 | 0.71 | 3.09 | 1.39 | 3.28 | 92.24 |
LibriSpeech | Lecture | 10.20 | 3.27 | 5.30 | 1.63 | 2.77 | 5.80 | 19.61 | 71.82 |