2014-06-20 - 版本历史

Zhaomy：/* GFbank */

2014-06-22T04:33:07Z

‎GFbank

Zhaomy：/* GFbank */

2014-06-22T04:32:36Z

‎GFbank

2014年6月22日 (日) 04:31 Zhaomy

2014-06-22T04:31:39Z

2014年6月20日 (五) 02:39 Cslt

2014-06-20T02:39:52Z

Cslt：以内容“==Resoruce Building== * release management combing done. == Leftover questions== * Asymmetric window: Great improvement on training set(WER 34% to 24%), however the im...”创建新页面

2014-06-20T02:37:47Z

以内容“==Resoruce Building== * release management combing done. == Leftover questions== * Asymmetric window: Great improvement on training set(WER 34% to 24%), however the im...”创建新页面

新页面

==Resoruce Building==
* release management combing done.

== Leftover questions==
* Asymmetric window: Great improvement on training set(WER 34% to 24%), however the improvement is lost on test.
* Multi GPU training: Error encountered
* Multilanguage training
* Investigating LOUDS FST.
* CLG embedded decoder plus online compiler.
* DNN-GMM co-training

== AM development ==

=== Sparse DNN ===
* GA-based block sparsity (+++++++)
* Paper revision done.

===Noise training===
:* Paper writing will be started this week

===GFbank===

* Running into Sinovoice 8k 1400 + 100 mixture training.
* GFbank 14 xEnt iteration completed:
Huawei disanpi BJ mobile 8k English data
FBank non-stream (MPE4) 20.44% 22.28% 24.36%
GFbank stream (MPE4) - - -
GFbank non-stream (MPE) - - -

===Multilingual ASR===

HW 30h (HW TR LM not involved) HW30h (HW TR LM involved)
FBank non-stream (MPE4) 22.23 21.38
Fbank stream (monolang) 21.64 20.72

GFbank stream (MPE4) - - -
GFbank non-stream (MPE) - - -

===Denoising & Farfield ASR===

* Replay may cause time delay. This should be solved by cross-correlation detection.
* Single-layer network with more hidden units. failed.
* Looks like the problem resides in large magnitude on output data.
* New recordings (one almost near mic & one far field 2 meters)

Original model:

xEnt model:
middle-field far-field
dev93 74.79 96.68
eval92 63.42 94.75

MPE model:

MPE adaptation:

middle-field far-field
dev93 63.71 94.84
eval92 52.67 90.45

===VAD===

* DNN-based VAD (7.49) showers much better performance than energy based VAD (45.74)
* 100 X n (n<=3) hidden units with 2 output units seem sufficient for VAD

===Scoring===

* Collect more data with human scoring to train discriminative models

===Embedded decoder===

FSA size:

threshold 1e-5 1e-6 1e-7 1e-8 1e-9
5k 480k 5.5M 44M - 1.1G
10k 731k 7M 61M
20k 1.2M 8.8M 78M(301M)

<pre>
600 X 4+800 AM, beam9:
150k 20k 10k 5k
WER 15.96 - - -
RT X 0.94 - -
</pre>

==LM development==

===Domain specific LM===

* Baiduzhidao + Weibeo extraction done with various thresholds
* Looks like the extracted text can improve to some extent, but the major change seems come from pre-pocessing.

* Check proportion of tags int HW 30 h data

==Word2Vector==

===W2V based doc classification===

* Full Gaussian based doc vector
:* represent each doc with a Gaussian distribution of the word vectors it involved.
:* using k-nn to conduct classification

mean Eur Distance KL distance baseline (NB with mean)

Acc (50dim) 81.84 79.65 69.7

==Semantic word tree==
:* First version based on pattern match done
:* Filter with query log
:* Further refinement with Baidu Baike hierarchy

==NN LM==

* Character-based NNLM (6700 chars, 7gram), 500M data training done.
:* Inconsistent pattern in WER were found on Tenent test sets
:* probably need to use another test set to do investigation.

* Investigate MS RNN LM training

@@ 第24行： / 第24行： @@
 * FBank/GFbank, stream/non-stream MPE completed:
                                     Huawei disanpi     BJ mobile   8k English data
 FBank non-stream (MPE4)             20.44%              22.28%      24.36%
 FBank stream (MPE1)             20.17%              22.50%      21.63%

@@ 第23行： / 第23行： @@
 * Running into Sinovoice 8k 1400 + 100  mixture training.
 * FBank/GFbank, stream/non-stream MPE completed:
                                     Huawei disanpi     BJ mobile   8k English data
 FBank non-stream (MPE4)             20.44%              22.28%      24.36%
 FBank stream (MPE1)             20.17%              22.50%      21.63%
 GFbank stream    (MPE4)           20.69%                22.84%       24.45%
 GFbank non-stream (MPE)             -                     -           -
 ===Multilingual ASR===

@@ 第30行： / 第30行： @@
 ===Multilingual ASR===
                                     HW 30h (HW TR LM not involved)     HW30h (HW TR LM involved)
 FBank non-stream (MPE4)             22.23                                   21.38
@@ 第36行： / 第37行： @@
 GFbank stream    (MPE4)             -                     -           -
 GFbank non-stream (MPE)             -                     -           -
 ===Denoising & Farfield ASR===
@@ 第46行： / 第49行： @@
 Original model:
 xEnt model:
                 middle-field    far-field
@@ 第60行： / 第64行： @@
      eval92      52.67          90.45
 ===VAD===
@@ 第76行： / 第81行： @@
 ===Embedded decoder===
 FSA size:
 threshold  1e-5    1e-6   1e-7   1e-8    1e-9
 k         480k    5.5M   44M     -      1.1G
 k        731k     7M    61M
 k        1.2M    8.8M   78M(301M)
+</pre>
 <pre>
@@ 第108行： / 第113行： @@
 :* using k-nn to conduct classification
               mean Eur Distance     KL distance    baseline (NB with mean)
 Acc (50dim)    81.84            79.65                  69.7
+</pre>
 ==Semantic word tree==