2014年6月6日 (五) 07:02 Cslt

2014-06-06T07:02:00Z

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

2014-06-06T06:05:37Z

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

新页面

==Resoruce Building==
* Release management has been started

== Leftover questions==
* Asymmetric window: Great improvement on training set(WER 34% to 24%), however the improvement is lost on test. Overfitting?
* 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 (+++++)

===Noise training===
:* All experiments completed.
:* Paper writing will be started this week

===GFbank===

* Test on Tencent database is done. Better performance observed than Fbank
* Equal-loudness pre-filter added, slightly better performance was obtained
* Running into Sinovoice 8k 1400 + 100 mixture training. 9 xEnt iteration completed.

===Multilingual ASR===

* Multilingual LM decoding
* non-tag bug investigation with some digital string recordings
* Revert to hanzi numbers

===English model===

<pre>

(state-gauss = 10000 100000, various LM, beam 13)

1. Shujutang 100h chi-eng 16k:

LM/AM | xEnt | mpe_1 | mpe_2 | mpe_3 | mpe_4 |
--------- --------- --------- --------- --------- ---------
wsj | 23.86 | 20.95 | 20.90 | 20.84 | 20.81 |
cmu | 22.22 | - | - | - | 18.83 |
giga | 21.77 | - | - | - | 18.61 |
armid | 20.45 | - | - | - | - |

2. Shujutang 100H chi-eng 8k:

LM/AM | xEnt | mpe_1 | mpe_2 | mpe_3 | mpe_4 |
--------- --------- --------- --------- --------- ---------
wsj | 26.27 | 23.63 | 23.14 | 22.93 | 23.00 |
cmu | 24.11 | - | - | - | 20.36 |
giga | 23.11 | - | - | - | 20.11 |
armid | - | - | - | - | - |

3. voxforge pure eng 16k:

LM/AM | xEnt | mpe_1 | mpe_2 | mpe_3 | mpe_4 |
--------- --------- --------- --------- --------- ---------
wsj | 21.38 | 24.89 | 24.50 | 23.31 | 23.13 |
cmu | 24.00 | - | - | - | 21.33 |
giga | 18.75 | - | - | - | 22.45 |
armid | - | - | - | - | - |

4. fisher pure eng 8k:
Not finish yet.
LM/AM | xEnt | mpe_1 | mpe_2 | mpe_3 | mpe_4 |
--------- --------- --------- --------- --------- ---------
wsj | 40.65 | 36.16 | 35.94 | 35.88 | 35.80 |
cmu | 35.07 | - | - | - | 31.16 |
giga | 41.18 | - | - | - | 36.23 |
armid | - | - | - | - | - |
</pre>

===Denoising & Farfield ASR===

* Investigating DAE model
* Kaldi-based MSE obj training toolkit preparation

===VAD===

* DNN-based VAD (7.49) showers much better performance than energy based VAD (45.74)
* Need to test small scale network (+)
:* 600-800 network
:* 100 X 4 + 2

===Scoring===

* Bug for the stream mode fixed

===Embedded decoder===

* word list graph test passed
* wlist2LG toolkit checked in
* Prepare to deliver Android compiler options (.mk)
* Interface design should be completed in one day
* Prepare HCLG for 20k LM, decoding on progress.

==LM development==

===Domain specific LM===

* Retrieve both Baidu & microblog
* PPL testing
* Need to check into gitLab.

===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3行： / 第3行： @@
 == Leftover questions==
-* Asymmetric window: Great improvement on training set(WER 34% to 24%), however the improvement is lost on test. Overfitting?
+* 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
@@ 第13行： / 第13行： @@
 === Sparse DNN ===
-* GA-based block sparsity (+++++)
+* GA-based block sparsity (++++++)
 ===Noise training===
 :* Paper writing will be started this week
 ===GFbank===
-* Test on Tencent database is done. Better performance observed than Fbank
+* Running into Sinovoice 8k 1400 + 100  mixture training.
-* Equal-loudness pre-filter added, slightly better performance was obtained
+* GFbank 14 xEnt iteration completed:
-* Running into Sinovoice 8k 1400 + 100  mixture training. 9 xEnt iteration completed.
+                                   Huawei disanpi     BJ mobile   8k English data
+FBank non-stream (17 iteration)     22.01%              26.63%     -
 ===Multilingual ASR===
 * Multilingual LM decoding
-* non-tag bug investigation with some digital string recordings
+* TAG-based decoding still problematic. Decoding goes into subgraph, however the decoding results are incorrect.
-* Revert to hanzi numbers
+* Investigate with free-loop grammar.
+* Non-tag test should be conducted on both Baidu & micro blob data
+* Should test the 8k shujutang data on the mixture model.
-===English model===
-−
-<pre>
-−
-(state-gauss = 10000 100000, various LM, beam 13)
-−
-. Shujutang 100h chi-eng 16k:
-−
-  LM/AM  |  xEnt   |  mpe_1  |  mpe_2  |  mpe_3  |  mpe_4  |
---------- --------- --------- --------- --------- ---------
-   wsj   |  23.86  |  20.95  |  20.90  |  20.84  |  20.81  |
-   cmu   |  22.22  |    -    |    -    |    -    |  18.83  |
-   giga  |  21.77  |    -    |    -    |    -    |  18.61  |
-  armid  |  20.45  |    -    |    -    |    -    |    -    |
-−
-−
-. Shujutang 100H chi-eng 8k:
-−
-  LM/AM  |  xEnt   |  mpe_1  |  mpe_2  |  mpe_3  |  mpe_4  |
---------- --------- --------- --------- --------- ---------
-   wsj   |  26.27  |  23.63  |  23.14  |  22.93  |  23.00  |
-   cmu   |  24.11  |    -    |    -    |    -    |  20.36  |
-   giga  |  23.11  |    -    |    -    |    -    |  20.11  |
-  armid  |    -    |    -    |    -    |    -    |    -    |
-−
-−
-. voxforge pure eng 16k:
-−
-  LM/AM  |  xEnt   |  mpe_1  |  mpe_2  |  mpe_3  |  mpe_4  |
---------- --------- --------- --------- --------- ---------
-   wsj   |  21.38  |  24.89  |  24.50  |  23.31  |  23.13  |
-   cmu   |  24.00  |    -    |    -    |    -    |  21.33  |
-   giga  |  18.75  |    -    |    -    |    -    |  22.45  |
-  armid  |    -    |    -    |    -    |    -    |    -    |
-−
-. fisher pure eng 8k:
-Not finish yet.
-  LM/AM  |  xEnt   |  mpe_1  |  mpe_2  |  mpe_3  |  mpe_4  |
---------- --------- --------- --------- --------- ---------
-   wsj   |  40.65  |  36.16  |  35.94  |  35.88  |  35.80  |
-   cmu   |  35.07  |    -    |    -    |    -    |  31.16  |
-   giga  |  41.18  |    -    |    -    |    -    |  36.23  |
-  armid  |    -    |    -    |    -    |    -    |    -    |
-</pre>
 ===Denoising & Farfield ASR===
-* Investigating DAE model
+* Add artificial reverberant with various energy decay & time delay. Draw a plot decay vs WER, delay vs WER.
-* Kaldi-based MSE obj training toolkit preparation
+* Use more training data to do adaptation.
 ===VAD===
@@ 第88行： / 第48行： @@
 * DNN-based VAD (7.49) showers much better performance than energy based VAD (45.74)
 * Need to test small scale network (+)
-:* 600-800 network
+:* 600-800 network test
-:* 100 X 4 + 2
+:* 100 X 4 + 2 network training
 ===Scoring===
-* Bug for the stream mode fixed
+* Collect more data with human scoring to train discriminative models
 ===Embedded decoder===
-* word list graph test passed
+X 4 + 10k AM:
-* wlist2LG toolkit checked in
-* Prepare to deliver Android compiler options (.mk)
-* Interface design should be completed in one day
-* Prepare HCLG for 20k LM, decoding on progress.
 ==LM development==
@@ 第110行： / 第69行： @@
 * Retrieve both Baidu & microblog
-* PPL testing
+* Need to check into gitLab(+).
 * Need to check into gitLab.
 ===NN LM===

2014-06-06 - 版本历史

2014年6月6日 (五) 07:02 Cslt

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