2013-04-12 - 版本历史

2013年4月19日 (五) 05:40 Liuc

2013-04-19T05:40:24Z

2013年4月19日 (五) 05:36 Liuc

2013-04-19T05:36:48Z

2013年4月12日 (五) 09:12 Wangd

2013-04-12T09:12:03Z

2013年4月12日 (五) 07:48 Liuc

2013-04-12T07:48:13Z

2013年4月12日 (五) 06:24 Wangd

2013-04-12T06:24:04Z

2013年4月12日 (五) 06:23 Wangd

2013-04-12T06:23:40Z

2013年4月12日 (五) 06:07 Wangd

2013-04-12T06:07:56Z

2013年4月12日 (五) 05:31 Wangd

2013-04-12T05:31:21Z

Wangd：以内容“1. Data sharing Done. 2. DNN progress Chao report. 3.Kaldi/HTK merge 4. Embedded progress (1). GFCC testing. noisy robust while not very good in silence. (2). P...”创建新页面

2013-04-12T05:26:14Z

以内容“1. Data sharing Done. 2. DNN progress Chao report. 3.Kaldi/HTK merge 4. Embedded progress (1). GFCC testing. noisy robust while not very good in silence. (2). P...”创建新页面

新页面

1. Data sharing

Done.

2. DNN progress

Chao report.

3.Kaldi/HTK merge

4. Embedded progress

(1). GFCC testing. noisy robust while not very good in silence.

(2). Prototype design. Application design is on going. Plan to deliver DNN decoder. Sparse DNN might be a good solution.

@@ 第9行： / 第9行： @@
   (1) 400 hour BN training is done. MFCC+LDA (300/1200/1200/1220/40/1200/38xx), followed by (MFCC+BN) with LDA.
   (2) comparision between MFCC and BN (fmpe applied). Relative improvement is 17.8% on chslm-biglm and 13.2% on chslm.
-  (3) BN system and hybrind system: relative perforamnce comparision:9.8% for hybrid, 6.6% for bn, 9.7% for bn+mfcc.
+  (3) BN system and hybrind system: relative perforamnce improvement comparision:9.8% for hybrid, 6.6% for bn, 9.7% for bn+mfcc.
   (4) GPU and CPU style comparision: still on progress. Working on data checking. SGE is still problematic (Chao can help). Hopefully done in 1 or two weeks.
   (5) RTF comparision between DNN hybrid and GMM: 0.57 vs 0.36.

@@ 第9行： / 第9行： @@
   (1) 400 hour BN training is done. MFCC+LDA (300/1200/1200/1220/40/1200/38xx), followed by (MFCC+BN) with LDA.
   (2) comparision between MFCC and BN (fmpe applied). Relative improvement is 17.8% on chslm-biglm and 13.2% on chslm.
-  (3) BN system and hybrind system: relative perforamnce comparision %xx vs %xx.
+  (3) BN system and hybrind system: relative perforamnce comparision:9.8% for hybrid, 6.6% for bn, 9.7% for bn+mfcc.
   (4) GPU and CPU style comparision: still on progress. Working on data checking. SGE is still problematic (Chao can help). Hopefully done in 1 or two weeks.
   (5) RTF comparision between DNN hybrid and GMM: 0.57 vs 0.36.

@@ 第11行： / 第11行： @@
   (3) BN system and hybrind system: relative perforamnce comparision %xx vs %xx.
   (4) GPU and CPU style comparision: still on progress. Working on data checking. SGE is still problematic (Chao can help). Hopefully done in 1 or two weeks.
-  (5) RTF comparision between DNN hybrid and GMM: 57% vs 36%.
+  (5) RTF comparision between DNN hybrid and GMM: 0.57 vs 0.36.

@@ 第8行： / 第8行： @@
   (1) 400 hour BN training is done. MFCC+LDA (300/1200/1200/1220/40/1200/38xx), followed by (MFCC+BN) with LDA.
-  (2) comparision between MFCC and BN (fmpe applied). Relative improvement is xx%.
+  (2) comparision between MFCC and BN (fmpe applied). Relative improvement is 17.8% on chslm-biglm and 13.2% on chslm.
   (3) BN system and hybrind system: relative perforamnce comparision %xx vs %xx.
   (4) GPU and CPU style comparision: still on progress. Working on data checking. SGE is still problematic (Chao can help). Hopefully done in 1 or two weeks.
-  (5) RTF comparision between DNN hybrid and GMM: %xx vs %xx.
+  (5) RTF comparision between DNN hybrid and GMM: 57% vs 36%.

←上一版本		2013年4月12日 (五) 06:24的版本
第26行：		第26行：

	(3). QA LM training, word files ready. trying to start the LM training. Refer to the doc Chao provided.		(3). QA LM training, word files ready. trying to start the LM training. Refer to the doc Chao provided.
		+
	/nfs/asrhome/asr/lm/chs.lm/lm.qa		/nfs/asrhome/asr/lm/chs.lm/lm.qa

@@ 第1行： / 第1行： @@
 . Data sharing
-  (1) Acoustic data ready. PLP feature + HLDA (with LDA transform). Model is PLP+HLDA+ML+MPE. All software ready. Good.
+  (1) Acoustic data ready. The feature is PLP with HLDA, and the model is PLP+HLDA+MPE. All softwares ready.
   (2)The LM data and model are being transferred.
@@ 第7行： / 第7行： @@
 . DNN progress
-  (1) 400hour Tencent data training ready. MFCC+LDA (300/1200/1200/1220/40/1200/38xx), followed by (MFCC+BN) X LDA.
+  (1) 400 hour BN training is done. MFCC+LDA (300/1200/1200/1220/40/1200/38xx), followed by (MFCC+BN) with LDA.
-  (2) comparision between MFCC and BN with fmpe. Relative improvement is xx%.
+  (2) comparision between MFCC and BN (fmpe applied). Relative improvement is xx%.
-  (3) BN system and hybrind system: relative perforamnce.
+  (3) BN system and hybrind system: relative perforamnce comparision %xx vs %xx.
-  (4) GPU and CPU style comparision: still on progress. Data checking. SGE is still problematic (Chao can help).
+  (4) GPU and CPU style comparision: still on progress. Working on data checking. SGE is still problematic (Chao can help). Hopefully done in 1 or two weeks.
-  (5) RTF comparision between DNN hybrid and GMM:
+  (5) RTF comparision between DNN hybrid and GMM: %xx vs %xx.
 .Kaldi/HTK merge
-  (1) HTK2Kaldi: problematic. HMM structure is erroratic. Need to modify the toolkit in Kaldi.
+  (1) HTK2Kaldi: The tool kaldi delivered is problematic. The HMM structure seems erroratic. Need to make correction (hopefully in 1 week).
-  (2) Kaldi2HTK: need to design a new tool.
+  (2) Kaldi2HTK: need to design a new tool (possibley in 1 week).
 . Embedded progress
-(1). GFCC testing. noisy robust while not very good in silence.
+(1). GFCC training/testing. The GFCC seems highly robust to noise,  while not as  good as MFCC in silence.
 (2). Prototype design. Application design is on going. Plan to deliver DNN decoder. Sparse DNN might be a good solution.

←上一版本		2013年4月12日 (五) 05:31的版本
第8行：		第8行：

	3.Kaldi/HTK merge		3.Kaldi/HTK merge
		+
		+	Chao report.
		+

	4. Embedded progress		4. Embedded progress