since loss reduce more when loss div batchsize, less lm alpha can be better.

5 years ago · 02e63b060c
parent e520077652
commit 02e63b060c
2 changed files with 2 additions and 2 deletions
--- a/examples/aishell/README.md
+++ b/examples/aishell/README.md
@ -3,5 +3,5 @@
 ## CTC
 | Model | Config | Test Set |  CER | Valid Loss |
 | --- | --- | --- | --- | --- |
-| DeepSpeech2 | conf/deepspeech2.yaml | test | 0.078786 | 7.036566 |
+| DeepSpeech2 | conf/deepspeech2.yaml | test | 0.078242 | 7.036566 |
 | DeepSpeech2 | release 1.8.5 | test | 0.080447 | - |
--- a/examples/aishell/conf/deepspeech2.yaml
+++ b/examples/aishell/conf/deepspeech2.yaml
@ -39,7 +39,7 @@ decoding:
  error_rate_type: cer 
  decoding_method: ctc_beam_search
  lang_model_path: data/lm/zh_giga.no_cna_cmn.prune01244.klm
-  alpha: 2.6
+  alpha: 2.5
  beta: 5.0
  beam_size: 300
  cutoff_prob: 0.99