cslt Wiki - 用户贡献 [zh-cn]

2016-06

2016-06-13T02:34:07Z

Zxw：/* 总结 */

'''警告：'''“2016-06”指向这里，但您没有足够的权限来访问它。

2016-06

2016-06-13T02:14:22Z

Zxw：/* 总结 */

'''警告：'''“2016-06”指向这里，但您没有足够的权限来访问它。

2016-06

2016-06-13T02:13:09Z

Zxw：/* list2 */

'''警告：'''“2016-06”指向这里，但您没有足够的权限来访问它。

2016-06

2016-06-13T02:11:51Z

Zxw：/* list1 */

'''警告：'''“2016-06”指向这里，但您没有足够的权限来访问它。

2016-06

2016-06-13T02:07:52Z

Zxw：/* list1 */

'''警告：'''“2016-06”指向这里，但您没有足够的权限来访问它。

2016-06

2016-06-12T07:27:25Z

Zxw：/* list1 */

'''警告：'''“2016-06”指向这里，但您没有足够的权限来访问它。

2016-06

2016-06-12T05:23:34Z

Zxw：以“=list1= {| class="wikitable" |+ 日常花费 ! 时间 !! 事项 !! 支出(元) !!　详细 !! 支付 !!经手人 |- !2016-05-24!! 同利节老师聚餐 !! 315 !! !!...”为内容创建页面

'''警告：'''“2016-06”指向这里，但您没有足够的权限来访问它。

Cost

2016-06-12T05:15:03Z

Zxw：

'''警告：'''“Cost”指向这里，但您没有足够的权限来访问它。

Xuewei Zhang 2016-06-06

2016-06-06T00:47:22Z

Zxw：以“*restaurant training :* revise lexicon and training tag LM text :* test beamforming package library :* Before training adaptation, process data and label sounds”为内容创建页面

*restaurant training
:* revise lexicon and training tag LM text
:* test beamforming package library
:* Before training adaptation, process data and label sounds

2016-06-06

2016-06-06T00:43:17Z

Zxw：

[[Qixin Wang 2016-06-06]]

[[Maoning Wang 2016-06-06]]

[[Lantian Li 2016-06-06]]

[[Zhiyuan Tang 2016-06-06]]

[[Xuewei Zhang 2016-06-06]]

Huilian-cslt-week-2016-6-2

2016-06-02T02:27:10Z

Zxw：

==Environment==

==RNN Poem Processing==
* Theme training --valid set converge
:* segment word
:* optimize speed done
:* theme poem generation
:* Wechat version to Huilian

==RNN Writing Processing==
* neural model learning transform --Xingchao,Yanxu
:* add attention
:* use tensor flow
:* LSTM --Yanxu

==RNN QA(release in June)==
* attention based QA (Solution for template expansion)
* train chatting model
:* RNN generateion version to Huilian(2016.6.5)
* CDSSM vector --Xingchao
* attention and memory QA
:* hold

Huilian-cslt-week-2016-6-2

2016-06-02T01:40:17Z

Zxw：以“==Environment== ==RNN Poem Processing== * Theme training --valid set converge :* segment word :* optimize speed done :* theme poem generation ==RNN Writing P...”为内容创建页面

==Environment==

==RNN Poem Processing==
* Theme training --valid set converge
:* segment word
:* optimize speed done
:* theme poem generation

==RNN Writing Processing==
* neural model learning transform --Xingchao,Yanxu
:* add attention
:* use tensor flow
:* LSTM --Yanxu

==RNN QA(release in June)==
* attention based QA (Solution for template expansion)
* train chatting model
* CDSSM vector --Xingchao
* attention and memory QA
:* hold

Huilan-cstl-lab-status

2016-06-02T01:39:51Z

Zxw：

Xuewei Zhang 2016-05-30

2016-05-30T00:52:37Z

Zxw：

*restaurant training
:*add tag to LM using text in training tag LM set WER:8.94%
:*recording
:*test using new recording data by adapting the weight of adding tag(0.02) and merging LM(-3) WER:14.45%

Xuewei Zhang 2016-05-30

2016-05-30T00:51:51Z

Zxw：以“*restaurant training :*add tag to LM using text in training tag LM set(8.94%) :*recording :*test using new recording data (14.45% by adapting the weight of adding ta...”为内容创建页面

*restaurant training
:*add tag to LM using text in training tag LM set(8.94%)
:*recording
:*test using new recording data (14.45% by adapting the weight of adding tag(0.02) and merging LM(-3))

2016-05-30

2016-05-30T00:42:29Z

Zxw：

[[Mengyuan Zhao 2016-05-30]]

[[Maoning Wang 2016-05-30]]

[[Aodong Li 2016-05-30]]

[[Xuewei Zhang 2016-05-30]]

Sinovoice-2016-5-26

2016-05-26T06:30:07Z

Zxw：

==Data==

*16K LingYun
:* 2000h data ready
:* 4300h real-env data to label

* YueYu
:* Total 250h(190h-YueYu + 60h-English)
:* Add 60h YueYu
:* CER: 75%->76%

* WeiYu
:* 8k more data
:* 50h for training
:* 120h labeled ready

* PingAn
:*100h User data done

==Model Training==
===Deletion Error Problem===
* Add one noise phone to alleviate the silence over-training, looks OK.
* Omit sil accuracy in discriminative training
* H smoothing of XEnt and MPE, no significant affect.
* Add one silence arc from start-state to end-state

===Big-Model Training===
====16k====
* Done!
====8k=====

=====Project=====
* PingAn

=========================================================================================
| AM / config | all | KeHu wer || KeHu no-ins |
-----------------------------------------------------------------------------------------
| tdnn 7-2048 xEnt | 16.45 | 36.49 || 25.18 |
| tdnn 7-2048 MPE | 15.22 | 32.77 || 23.48 |
| tdnn 7-2048 MPE adapt-PABX | 14.67 | 31.33 || 22.76 |
-----------------------------------------------------------------------------------------
| tdnn 7-1024 xEnt | 16.60 | 35.91 || 25.58 |
| tdnn 7-1024 MPE 2e-6 | 15.67 | 32.77 || 26.09 |
| tdnn 7-1024 MPE 2e-5 1.mdl | 15.54 | 32.77 || 26.29 |
| tdnn 7-1024 MPE 1e-5 4.mdl | 15.76 | 33.55 || 27.20 |
| tdnn 7-1024 MPE adapt-PABX | 14.80 | 30.48 || 22.56 |
-----------------------------------------------------------------------------------------
| spn 7-1024 xEnt | 16.49 | 36.23 || 24.59 |
| spn 7-1024 xEnt xEnt-PA_user 101.mdl| 16.19 | 33.22 || 22.69 |
| spn 7-1024 xEnt xEnt-PA_user mpe | 15.24 | 32.77 || 21.65 |
| spn 7-1024 MPE-1000H 23.mdl | 15.29 | 33.09 || 21.65 |
| spn 7-1024 MPE adapt-PA_all 29.mdl | 15.11 | 33.42 || 21.84 |
| spn 7-1024 MPE adapt-PA_user 2e-5 | 15.31 | 31.79 || 20.14 |
| spn 7-1024 MPE adapt-PA_user Hs 2e-5| 15.32 | 32.24 || 20.93 |
=========================================================================================

=====================================================================================================
| LM / config | KeHu |KeHu check_zxm_recheck| KeHu final |
-----------------------------------------------------------------------------------------------------
| baseline | 20.14 | 19.40 | 18.26 |
-----------------------------------------------------------------------------------------------------
|bank+baoxian.chart+word.w0.9 | - | 19.27 | 17.88 |
|bank+baoxian+guojiadianwang.chart+word.w0.9| - | 19.02 | 17.94 |
|bank+baoxian+guojiadianwang_w0.9 | - | 19.02 | - |
|bank+baoxian_w0.9 | - | 19.20 | - |
|baoxian+bank_w0.9 | - | 19.02 | - |
|baoxian+user200h.chart+word.w0.9 | - | 19.20 | - |
|baoxian+user200h_w0.8 | - | 19.02 | - |
|baoxian+user200h_w0.9 | - | 19.08 | 18.01 |
|baoxian+user200h.w0.9w0.9 | - | 19.08 | - |
|baoxian+user200h.chart1e-7.w0.9w0.1 | - | - | 23.82 |
|baoxian+user200h.chart1e-7.w0.9w0.9 | - | - | 18.26 |
=====================================================================================================

* LiaoNingYiDong:
=========================================================================
| AM / config | LNYD | LNYD re-tag |
-------------------------------------------------------------------------
| tdnn 7-2048 xEnt | 21.51 | |
| tdnn 7-2048 MPE | 20.09 | |
| tdnn 7-2048 MPE adapt-LNYD | 17.92 | 16.29 |
-------------------------------------------------------------------------
| tdnn 7-1024 xEnt | 21.72 | |
| tdnn 7-1024 MPE | 20.99 | |
| cnn 7-1024 xEnt 600.mdl | 21.03 | |
| cnn 7-1024 MPE 12.mdl | 19.80 | |
| cnn 7-1024 MPE adapt-LNYD 41.mdl | 17.96 | 15.93 |
-------------------------------------------------------------------------
| spn 7-1024 xEnt | 21.70 | |
| spn 7-1024 MPE-1000H 23.mdl | 19.97 | |
| spn 7-1024 MPE adapt-LNYD | 18.67 | |
| spn cnn 7-1024 xEnt 300.mdl | 22.26 | |
========================================================================

===Embedding===
* The size of nnet1 AM is 6.4M (3M after decomposition). So we need to control AM size within 10M.
* 5*500-2400 TDNN no-svd/svd100 model, MPE training done

LM=1e-5, beam=9, max-active=5000
=============================================================================================================
| AM / testset | test_1000ju | test_2000ju | test_8000ju | test_10000ju |
-------------------------------------------------------------------------------------------------------------
| nnet1 4*600+800 xEnt (6.4M) | 25.30 | 40.48 | | |
| nnet1 4*600+800 mpe (6.4M) | 20.75 | 35.33 | | |
-------------------------------------------------------------------------------------------------------------
| nnet3 5*500 mpe (13M) | 16.18 | 29.53 | | |
| nnet3 5*500 svd-100 mpe (9.5M) | 17.69 | 30.11 | | |
=============================================================================================================

===Character LM===
* Except Sogou-2T, 9-gram has been done.
* Add word boundary tag to Character-LM trainig done
:* 9-gram
:* Except Weibo & Sogou-2T
:* 1e-7(13M) wer17.91 compared with 1e-7(no-boundary,71M) 13.4
:* 1e-8(54M) wer17.54

* Prepare specific domain vocabulary
:* Dianxin/Baoxian/Dianli

*DT lm training
:* ReFr

*Merge Character-LM & word-LM
:* Union
:* Compose, success.
* 2-step decoding: first, character-based LM. Then, word-based LM.

===Problem===
* Pingan & Yueyu too much deletion error.
:* TDNN deletion error rate > DNN deletion error rate
:* TDNN Silence scale is too sensitive for different test cases.
* cmvn causes performance reduction.

==SiaSun Robot==
* Beam-forming algorithm test
* NN-model based beam-forming

==SID==
===Digit===
* Engine Package

Huilian-cslt-week-2016-5-26

2016-05-26T01:55:08Z

Zxw：

Huilian-cslt-week-2016-5-26

2016-05-26T01:36:39Z

Zxw：以“==Environment== ==RNN Poem Processing== * Theme training --valid set converge :* segment word :* optimize speed done :* theme poem generation ==RNN Writing Pr...”为内容创建页面

==Environment==

==RNN Poem Processing==
* Theme training --valid set converge
:* segment word
:* optimize speed done
:* theme poem generation

==RNN Writing Processing==
* hold
* neural model learning transform --Xingchao
:* add attention
:* use tensor flow
* BPL --Yanxu

==RNN QA(release in June)==
* attention based QA (Solution for template expansion)--Tianyi
:* collect Simsimi corpus --Bingdong
::* train chatting model
* RNN vector without attention --Tianyi
:* DSSM vector --Xingchao
:* compare to DSSM vector
:* attention and memory QA

Huilan-cstl-lab-status

2016-05-26T01:36:14Z

Zxw：

Xuewei Zhang 2016-05-23

2016-05-23T00:43:04Z

Zxw：以“*restaurant training :* add tag to LM ::* add one tag to make one graph 24.28 ::* add all tags to make one graph”为内容创建页面

*restaurant training
:* add tag to LM
::* add one tag to make one graph 24.28
::* add all tags to make one graph

2016-05-23

2016-05-23T00:40:34Z

Zxw：

[[Lantian Li 2016-05-23]]

[[Maoning Wang 2016-05-23]]

[[Zhiyong Zhang 2016-05-23]]

[[Zhiyuan Tang 2016-05-23]]

[[Xuewei Zhang 2016-05-23]]

Status report

2016-05-16T00:54:25Z

Zxw：

[[2016-01-04]]

[[2016-01-11]]

[[2016-01-18]]

[[2016-01-25]]

[[2016-02-01]]

[[2016-02-22]]

[[2016-02-29]]

[[2016-03-07]]

[[2016-03-14]]

[[2016-03-21]]

[[2016-03-28]]

[[2016-04-04]]

[[2016-04-11]]

[[2016-04-18]]

[[2016-04-25]]

[[2016-05-02]]

[[2016-05-09]]

[[2016-05-16]]

Xuewei Zhang 2016-05-16

2016-05-16T00:37:20Z

Zxw：以“*restaurant training :* cnn tdnn dae training :* beamforming + dae training”为内容创建页面

*restaurant training
:* cnn tdnn dae training
:* beamforming + dae training

2016-05-16

2016-05-16T00:35:42Z

Zxw：

[[Jingyi Lin 2016-05-16]]

[[Qixin Wang 2016-05-16]]

[[Lantian Li 2016-05-16]]

[[Mengyuan Zhao 2016-05-16]]

[[Aodong Li 2016-05-16]]

[[Yang Wang 2016-05-16]]

[[Xuewei Zhang 2016-05-16]]

Xuewei Zhang 2016-05-09

2016-05-09T00:48:49Z

Zxw：

*restaurant training
:*segment audio data
:*beamforming test(has problem)
:*cnn_tddn dae [http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?step=view_request&cvssid=535 here]
::* paste 4 channels feats to train and adjust the parameters
::* mix different channel data to train
:* conpute SNR
::* before dinner SNR is 5dB
::* after dinner SNR is 14dB

Xuewei Zhang 2016-05-09

2016-05-09T00:48:33Z

Zxw：

*restaurant training
:*segment audio data
:*beamforming test(has problem)
:*cnn_tddn dae [ http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?step=view_request&cvssid=535 here]
::* paste 4 channels feats to train and adjust the parameters
::* mix different channel data to train
:* conpute SNR
::* before dinner SNR is 5dB
::* after dinner SNR is 14dB

Xuewei Zhang 2016-05-09

2016-05-09T00:46:35Z

Zxw：

*restaurant training
:*segment audio data
:*beamforming test(has problem)
:*cnn_tddn dae http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?step=view_request&cvssid=535
::* paste 4 channels feats to train and adjust the parameters
::* mix different channel data to train
:* conpute SNR
::* before dinner SNR is 5dB
::* after dinner SNR is 14dB

Xuewei Zhang 2016-05-09

2016-05-09T00:46:19Z

Zxw：

*restaurant training
:*segment audio data
:*beamforming test(has problem)
:*cnn_tddn dae [here http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?step=view_request&cvssid=535]
::* paste 4 channels feats to train and adjust the parameters
::* mix different channel data to train
:* conpute SNR
::* before dinner SNR is 5dB
::* after dinner SNR is 14dB

Xuewei Zhang 2016-05-09

2016-05-09T00:44:20Z

Zxw：

*restaurant training
:*segment audio data
:*beamforming test(has problem)
:*cnn_tddn dae
::* paste 4 channels feats to train and adjust the parameters
::* mix different channel data to train
:* conpute SNR
::* before dinner SNR is 5dB
::* after dinner SNR is 14dB

Xuewei Zhang 2016-05-09

2016-05-09T00:42:21Z

Zxw：

*restaurant training
:*segment audio data
:*beamforming test(has problem)
:*cnn_tddn dae
::* paste 4 channels feats to train and adjust the parameters
::* mix different channel data to train

Xuewei Zhang 2016-05-09

2016-05-09T00:42:05Z

Zxw：以“*restaurant traiining :*segment audio data :*beamforming test(has problem) :*cnn_tddn dae ::* paste 4 channels feats to train and adjust the parameters ::* mix dif...”为内容创建页面

*restaurant traiining
:*segment audio data
:*beamforming test(has problem)
:*cnn_tddn dae
::* paste 4 channels feats to train and adjust the parameters
::* mix different channel data to train

2016-05-09

2016-05-09T00:37:45Z

Zxw：

[[Mengyuan Zhao 2016-05-09]]

[[Lantian Li 2016-05-09]]

[[Yang Wang 2016-05-09]]

[[Xuewei Zhang 2016-05-09]]

2016-05-09

2016-05-09T00:37:36Z

Zxw：

[[Mengyuan Zhao 2016-05-09]]

[[Lantian Li 2016-05-09]]

[[Yang Wang 2016-05-09]]

[[Xuewei Zhang 2016-05-09]]

Huilian-cslt-week-2016-4-28

2016-04-28T02:38:50Z

Zxw：

Huilian-cslt-week-2016-4-28

2016-04-28T01:51:23Z

Zxw：以“==Environment== ==RNN Poem Processing== * Theme training --valid set converge :* segment word :* optimize speed done ==RNN Writing Processing== * hold * neu...”为内容创建页面

==Environment==

==RNN Poem Processing==
* Theme training --valid set converge
:* segment word
:* optimize speed done

==RNN Writing Processing==
* hold
* neural model learnig transform --Xingchao
:* add attention
:* use tensor flow
* BPL --Yanxu

==RNN QA(release in June)==
* attention based QA (Solution for template expansion)--Tianyi
:* collect Simsimi corpus --Bingdong
::* train chatting model
* RNN vector without attention --Tianyi
:* DSSM vector --Xingchao
:* compare to DSSM vector
:* attention and memory QA

Huilan-cstl-lab-status

2016-04-28T01:50:39Z

Zxw：

Xuewei Zhang 2016-04-25

2016-04-25T00:47:55Z

Zxw：以“*beamforming test :*reproduce MVDR beamforming done :*add postfilter done ;*generate data according to delay between source and microphones :* segment audio data”为内容创建页面

*beamforming test
:*reproduce MVDR beamforming done
:*add postfilter done
;*generate data according to delay between source and microphones
:* segment audio data

2016-04-25

2016-04-25T00:40:13Z

Zxw：

[[Tianyi Luo 2016-04-25]]

[[Mengyuan Zhao 2016-04-25]]

[[Zhiyuan Tang 2016-04-25]]

[[Hui Tang 2016-04-25]]

[[Yang Wang 2016-04-25]]

[[Xuewei Zhang 2016-04-25]]

Xuewei Zhang 2016-04-18

2016-04-18T00:55:39Z

Zxw：

* beamforming test
:* reproduce DS beamforming (delay and sum beamforming)
:* reproduce SD beamforming (super-directive beamforming)
:* adjust the distance of microphones or the location of source,test WER compared to WER from Jietong beamforming method [http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?account=zxw&step=view_request&cvssid=529 here]
* record in meeting room
* read papers
:* HYBRID ACOUSTIC MODELS FOR DISTANT AND MULTICHANNEL LARGE VOCABULARY SPEECH RECOGNITION
:* FACTORED SPATIAL AND SPECTRAL MULTICHANNEL RAW WAVEFORM CLDNNS

Xuewei Zhang 2016-04-18

2016-04-18T00:52:02Z

Zxw：以“* beamforming test :* reproduce DS beamforming (delay and sum beamforming) :* reproduce SD beamforming (super-directive beamforming) :* adjust the distance of micro...”为内容创建页面

* beamforming test
:* reproduce DS beamforming (delay and sum beamforming)
:* reproduce SD beamforming (super-directive beamforming)
:* adjust the distance of microphones or the location of source,test WER [http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?account=zxw&step=view_request&cvssid=529 here]
* record in meeting room
* read papers
:* HYBRID ACOUSTIC MODELS FOR DISTANT AND MULTICHANNEL LARGE VOCABULARY SPEECH RECOGNITION
:* FACTORED SPATIAL AND SPECTRAL MULTICHANNEL RAW WAVEFORM CLDNNS

2016-04-18

2016-04-18T00:44:59Z

Zxw：

[[Tianyi Luo 2016-04-18]]

[[Hui Tang 2016-04-18]]

[[Mengyuan Zhao 2016-04-18]]

[[Xiangyu Zeng 2016-04-18]]

[[Zhiyuan Tang 2016-04-18]]

[[Zhiyong Zhang 2016-04-18]]

[[Chao Xing 2016-04-18]]

[[Jiyuan Zhang 2016-04-18]]

[[Xuewei Zhang 2016-04-18]]

Huilian-cslt-week-2016-4-14

2016-04-14T01:56:36Z

Zxw：以“==Environment== ==RNN Poem Processing== * Theme training --valid set converge :* segment word :* optimize speed done ==RNN Writing Processing== * hold * neu...”为内容创建页面

Huilan-cstl-lab-status

2016-04-14T01:31:45Z

Zxw：

XueweiZhang 2016-04-11

2016-04-11T00:39:22Z

Zxw：以“*Demo :* make a demo based on Winxin”为内容创建页面

*Demo
:* make a demo based on Winxin

2016-04-11

2016-04-11T00:37:59Z

Zxw：

[[Mengyuan Zhao 2016-04-11]]

[[Lantian Li 2016-04-11]]

[[Hui Tang 2016-04-11]]

[[Chao Xing 2016-04-11]]

[[Yang Wang 2016-04-11]]

[[Zhiyong Zhang 2016-04-11]]

[[XueweiZhang 2016-04-11]]

Huilian-cslt-week-2016-4-7

2016-04-07T01:57:52Z

Zxw：/* RNN QA(release in June) */

==Environment==

==RNN Poem Processing==
* Theme training --valid set converge
:* segment word
:* optimize speed done
* automatic music generation --Jiyuan
:* generate piano
:* generate Chinese music(Lijun Deng)

==RNN Writing Processing==
* neural model learnig transform --Xingchao
:* add attention
:* use tensor flow
* BPL --Yanxu

==RNN QA(release in June)==
* attention based QA (Solution for template expansion)--Tianyi
:* collect Simsimi corpus --Bingdong
::* train chatting model
* RNN vector without attention --Tianyi
:* DSSM vector --Xingchao
:* compare to DSSM vector
:* attention and memory QA

Huilian-cslt-week-2016-4-7

2016-04-07T01:52:05Z

Zxw：/* RNN QA(release June) */

Huilian-cslt-week-2016-4-7

2016-04-07T01:51:51Z

Zxw：/* RNN QA */

==Environment==

==RNN Poem Processing==
* Theme training --valid set converge
:* segment word
:* optimize speed done
* automatic music generation --Jiyuan
:* generate piano
:* generate Chinese music(Lijun Deng)

==RNN Writing Processing==
* neural model learnig transform --Xingchao
:* add attention
:* use tensor flow
* BPL --Yanxu

==RNN QA(release June)==
* attention based QA (Solution for template expansion)--Tianyi
:* collect Simsimi corpus --Bingdong
::* train chatting model
* RNN vector without attention --Tianyi
:* compare to DSSM vector
:* attention and memory QA

Huilian-cslt-week-2016-4-7

2016-04-07T01:48:59Z

Zxw：

==Environment==

==RNN Poem Processing==
* Theme training --valid set converge
:* segment word
:* optimize speed done
* automatic music generation --Jiyuan
:* generate piano
:* generate Chinese music(Lijun Deng)

==RNN Writing Processing==
* neural model learnig transform --Xingchao
:* add attention
:* use tensor flow
* BPL --Yanxu

==RNN QA==
* attention based QA (Solution for template expansion)--Tianyi
:* collect Simsimi corpus --Bingdong
::* train chatting model
* RNN vector without attention --Tianyi
:* compare to DSSM vector
:* attention and memory QA