2013-09-06 - 版本历史

Wangd：/* FBank features */

2013-09-06T06:36:01Z

‎FBank features

2013-09-06T06:30:21Z

‎Discriminative DNN

2013-09-06T06:29:35Z

‎Tencent exps

2013-09-06T06:29:05Z

‎Noisy training

2013-09-06T06:28:52Z

‎Stream decoding

2013-09-06T06:22:11Z

2013-09-06T05:52:43Z

‎FBank features

2013-09-06T05:52:13Z

‎FBank features

2013-09-06T05:51:00Z

‎Sparse DNN

2013-09-06T05:50:47Z

‎Sparse DNN

@@ 第60行： / 第60行： @@
 * FB feature is much better than both MFCC and GFCC. Probably due to the less information lost without DCT.
 * We need to investigate how many FBs are the most appropriate.
 * Inspired by the assumption of information lost with DCT, we need to test how another transform,  LDA,  leads to the similar information lost. We need to investigate which is the suitable dimension number for the LDA. We need to investigate non-linear discriminative approach which is simple but leads to less information lost.

←上一版本		2013年9月6日 (五) 06:30的版本
第4行：		第4行：

	== DNN progress ==		== DNN progress ==
−
−	~~=== Discriminative DNN ===~~
−
−	* 1000 hour training not yet finished.

	=== Sparse DNN ===		=== Sparse DNN ===

@@ 第71行： / 第71行： @@
 === Tencent exps ===
+N/A
 ==DNN Confidence estimation==

@@ 第86行： / 第86行： @@
 http://cslt.riit.tsinghua.edu.cn/cgi-bin/cvss/cvss_request.pl?account=wangd&step=view_request&cvssid=118
-{|calss='wikitable'
+{|class='wikitable'
 !    SNR  !!        MFCC     !!  GFCC
 |-

@@ 第105行： / 第105行： @@
 * The interface for server-side is done. For embedded-side is on development.
-* Fixed a bug which prompts intermediate results when silence encountered.
+* Fixed a bug which prompts intermediate results when short pause encountered.
 * Fixed a CMN bug for the last segment.

@@ 第42行： / 第42行： @@
 Test on 100 hour data, clean & noise speech.
 ,MFCC 100_1200_1200_1200_1200_3580
         map: %WER 23.75 [ 3474 / 14628, 134 ins, 373 del, 2967 sub ]

@@ 第39行： / 第39行： @@
 === FBank features ===
 ,MFCC 100_1200_1200_1200_1200_3580
@@ 第98行： / 第100行： @@
      speedup: %WER 64.78 [ 3404 / 5255, 13 ins, 1084 del, 2307 sub ]
 * Need to investigate Gammatone filter banks. The same idea as FB, that we want to keep the information as much as possible. And it is possible to combine FB and GFB to pursue a better performance.
-−
 === Tencent exps ===

@@ 第26行： / 第26行： @@
 |general ||34.41||34.34
 |-
-|online1 ||33.02||||32.92
+|online1 ||33.02||32.92
 |-
 |online2 ||25.99||26.06

@@ 第17行： / 第17行： @@
 |-
 |map     ||23.75 ||23.90
 |2044    ||21.47 ||21.45
 |notetp3 ||13.17 ||13.65
 |record1900 ||8.10|| 8.18
 |general ||34.41||34.34
 |online1 ||33.02||||32.92
 |online2 ||25.99||26.06
 |speedup ||23.52||23.58
 |}