Add results

Author: hirofumi0810

README.md

@@ -13,11 +13,12 @@ make KALDI=/path/to/kaldi TOOL=/path/to/save/tools
 ### Corpus
   - ASR
     - AISHELL-1
+    - AISHELL-2
     - AMI
     - CSJ
     - LaboroTVSpeech
     - Librispeech
-    - Switchboard (+ Fisher)
+    - Switchboard (+Fisher)
     - TEDLIUM2/TEDLIUM3
     - TIMIT
     - WSJ
@@ -61,6 +62,7 @@ make KALDI=/path/to/kaldi TOOL=/path/to/save/tools
     - Deep fusion [[link](https://arxiv.org/abs/1503.03535)]
     - Forward-backward attention decoding [[link](https://www.isca-speech.org/archive/Interspeech_2018/abstracts/1160.html)]
     - Ensemble decoding
+    - internal LM estimation [[link](https://arxiv.org/abs/2011.01991)]
   - Attention type
     - location-based
     - content-based
@@ -108,6 +110,11 @@ Multi-task learning (MTL) with different units are supported to alleviate data s
 | Transformer   | 5.0 | 5.4  |
 | Streaming MMA | 5.5 | 6.1  |
 
+### AISHELL-2 (CER)
+| Model         | test_android | test_ios | test_mic |
+| -----------   | ------------ | -------- | -------- |
+| Conformer LAS | 6.1          | 5.5      | 5.9      |
+
 ### CSJ (WER)
 | Model          | eval1 | eval2 | eval3 |
 | -------------- | ----- | ----- | ----- |

examples/aishell2/s5/RESULTS.md

@@ -0,0 +1,13 @@
+### Conformer-LAS + SpecAugment (no LM), hierarchical subsample1/8
+- conf: `conf/asr/conformer_kernel15_clamp10_hie_subsample8_las_ln_large.yaml`
+- decoding parameters
+  - epoch: 30
+  - n_average: 10
+  - beam width: 10
+  - lm_weight: 0.0
+
+| Eval Set | # Snt | # Wrd | Corr | Sub | Del | Ins | Err | S.Err |
+| -------- | ----- | ----- | ---- | --- | --- | --- | --- | ----- |
+|test_android|5000|49532|94.0|5.8|0.2|0.1|**6.1**|36.7|
+|test_ios|5000|49532|94.6|5.2|0.2|0.1|**5.5**|34.1|
+|test_mic|5000|49532|94.3|5.6|0.2|0.1|**5.9**|35.7|