Table 9 A comparison of using different bias factors and SAD dilation-erosion factors on ASR Model 2. ATWV values are evaluated on the SAD segmented dev dataset. “G” and “C” represent GMM-HMM aligner and CTC aligner, respectively
From: Timestamp-aligning and keyword-biasing end-to-end ASR front-end for a KWS system
bias factor | SAD α 0.1 | SAD α 0.2 | SAD α 0.3 | SAD α 0.4 | SAD α 0.5 |
|---|---|---|---|---|---|
δ | G | G | G / C | G / C | G / C |
0 (baseline) | 0.5673 | 0.5783 | 0.5917/0.5593 | 0.5906/0.5542 | 0.5973/0.5656 |
1/4 | 0.5808 | 0.6026 | 0.5935/0.5778 | 0.6089/0.5786 | 0.5923/0.5699 |
1/3 | 0.5891 | 0.5919 | 0.6009/0.5414 | 0.5889/0.5498 | 0.5975/0.5707 |
1/2 | 0.5744 | 0.5847 | 0.5830/0.5402 | 0.5789/0.5389 | 0.5901/0.5617 |
1 | 0.5753 | 0.5899 | 0.5910/0.5710 | 0.6077/0.5706 | 0.5997/0.5611 |
2 | 0.5805 | 0.5801 | 0.5956/0.5543 | 0.5749/0.5415 | 0.5657/0.5288 |
4 | 0.5445 | 0.5751 | 0.5837/0.5525 | 0.5699/0.5271 | 0.5467/0.4965 |