Update README.md
Browse files
README.md
CHANGED
|
@@ -4,4 +4,62 @@ license: apache-2.0
|
|
| 4 |
|
| 5 |
# Omni-temporal Classification
|
| 6 |
We propose BTC/OTC to directly train an ASR system leveraging weak supervision, i.e., speech with non-verbatim transcripts. This is achieved by using a special token
|
| 7 |
-
to model uncertainties (i.e., substitution errors, insertion errors, and deletion errors) within the WFST framework during training.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
|
| 5 |
# Omni-temporal Classification
|
| 6 |
We propose BTC/OTC to directly train an ASR system leveraging weak supervision, i.e., speech with non-verbatim transcripts. This is achieved by using a special token
|
| 7 |
+
to model uncertainties (i.e., substitution errors, insertion errors, and deletion errors) within the WFST framework during training.
|
| 8 |
+
|
| 9 |
+
### Results (ctc-greedy-search)
|
| 10 |
+
<table>
|
| 11 |
+
<tr>
|
| 12 |
+
<td rowspan=2>Training Criterion</td>
|
| 13 |
+
<td colspan=2>ssl</td>
|
| 14 |
+
<td colspan=2>fbank</td>
|
| 15 |
+
</tr>
|
| 16 |
+
<tr>
|
| 17 |
+
<td>test-clean</td>
|
| 18 |
+
<td>test-other</td>
|
| 19 |
+
<td>test-clean</td>
|
| 20 |
+
<td>test-other</td>
|
| 21 |
+
</tr>
|
| 22 |
+
<tr>
|
| 23 |
+
<td>CTC</td>
|
| 24 |
+
<td>100.0</td>
|
| 25 |
+
<td>100.0</td>
|
| 26 |
+
<td>99.89</td>
|
| 27 |
+
<td>99.98</td>
|
| 28 |
+
</tr>
|
| 29 |
+
<tr>
|
| 30 |
+
<td>OTC</td>
|
| 31 |
+
<td>11.89</td>
|
| 32 |
+
<td>25.46</td>
|
| 33 |
+
<td>20.14</td>
|
| 34 |
+
<td>44.24</td>
|
| 35 |
+
</tr>
|
| 36 |
+
</table>
|
| 37 |
+
|
| 38 |
+
### Results (1best, blank_bias=-4)
|
| 39 |
+
<table>
|
| 40 |
+
<tr>
|
| 41 |
+
<td rowspan=2>Training Criterion</td>
|
| 42 |
+
<td colspan=2>ssl</td>
|
| 43 |
+
<td colspan=2>fbank</td>
|
| 44 |
+
</tr>
|
| 45 |
+
<tr>
|
| 46 |
+
<td>test-clean</td>
|
| 47 |
+
<td>test-other</td>
|
| 48 |
+
<td>test-clean</td>
|
| 49 |
+
<td>test-other</td>
|
| 50 |
+
</tr>
|
| 51 |
+
<tr>
|
| 52 |
+
<td>CTC</td>
|
| 53 |
+
<td>98.40</td>
|
| 54 |
+
<td>98.68</td>
|
| 55 |
+
<td>99.79</td>
|
| 56 |
+
<td>99.86</td>
|
| 57 |
+
</tr>
|
| 58 |
+
<tr>
|
| 59 |
+
<td>OTC</td>
|
| 60 |
+
<td>6.59</td>
|
| 61 |
+
<td>15.98</td>
|
| 62 |
+
<td>11.78</td>
|
| 63 |
+
<td>32.38</td>
|
| 64 |
+
</tr>
|
| 65 |
+
</table>
|