=Project name=
Speaker Recognition on Trivial events

=Project members=
Dong Wang, Miao Zhang, Xiaofei Kang, Lantian Li, Zhiyuan Tang

=Introduction=
Trivial events are ubiquitous in human to human conversations, e.g., cough, laugh and sniff. Compared to regular
speech, these trivial events are usually short and unclear,
thus generally regarded as not speaker discriminative and so
are largely ignored by present speaker recognition research.
However, these trivial events are highly valuable in some particular circumstances such as forensic examination, as they
are less subjected to intentional change, so can be used to
discover the genuine speaker from disguised speech.
In this project, we collect a trivial event speech database
and report speaker recognition results on the database, by both
human listeners and machines.
We want to find out:
(1) which type of trivial event conveys more speaker information;
(2) who, human or machine, is more apt to identify speakers from these trivial events.

=Speaker feature learning=

The discovery of the short-time property of speaker traits is the key step towards speech signal factorization, as
the speaker trait is one of the two main factors: the other is linguistic content that we have known for a long time
being short-time patterns.

The key idea of speaker feature learning is simply based on the idea of discriminating training speakers based on
short-time frames by deep neural networks (DNN), date back to 2014 by Ehsan et al.[2]. As shown below, the output of the DNN
involves the training speakers, and the frame-level speaker features are read from the last hidden layer. The
basic assumption here is: if the output of the last hidden layer can be used as the input feature of the
last hidden layer (a software regression classifier), these features should be speaker discriminative.

[[文件:Dnn-spk.png|500px]]

However, the vanilla structure of Ehsan et al. performs rather poor compared to the i-vector counterpart. One reason is
that the simple back-end scoring is based on average to derive the utterance-based representations (called d-vectors) , but
another reason is the vanilla DNN structure that does not consider much of the context and pattern learning. We therefore
proposed a CT-DNN model that can learn stronger speaker features. The structure is shown below[1]:

[[文件:Ctdnn-spk.png|500px]]

Recently, we found that an 'all-info' training is effective for learning features. Looking back to DNN and CT-DNN, although the features
read from last hidden layer are discriminative, but not 'all discriminative', because some discriminant info can be also impelemented
in the last affine layer. A better strategy is let the feature generation net (feature net) learns all the things of discrimination.
To achieve this, we discarded the parametric classifier (the last affine layer) and use the simple cosine distance to conduct the
classification. An iterative training scheme can be used to implement this idea, that is, after each epoch, averaging the speaker
features to derive speaker vectors, and then use the speaker vectors to replace the last hidden layer. The training will be then
taken as usual. The new structure is as follows[4]:

[[文件:fullinfo-spk.png|500px]]

=Database design=

To collect the data, we designed a mobile application and
distributed it to people who agreed to participate. The application asked the participants to utter 6 types of trivial events
in a random order, and each event occurred 10 times randomly. The random order ensures a reasonable variance of the
recordings for each event. The sampling rate of the recordings was set to 16 kHz and the precision of the samples was
16 bits.

We first designed CSLT-TRIVIAL-I which involves 6 trivial events of speech, i.e., cough, laugh, 'hmm', 'tsk-tsk', 'ahem'and sniff.

[[文件:Cslt-trivial-1.png|500px]]

A speech database CSLT-DISGUISE-I of normal & disguised pairs was also designed.

[[文件:Cslt-disguise-1.png|500px]]

==Data downloading==

CSLT-TRIVIAL-I CSLT-DISGUISE-I are free for research. LICENSE is needed. Contact Dr. Dong Wang (wangdong99@mails.tsinghua.edu.cn).
Download link: https://github.com/CSLT-THU/TRIVIAL-EVENTS-RECOGNITION

=Human performance=

Human performed not well on CSLT-TRIVIAL-I, also on CSLT-DISGUISE-I with detection error rate (DER) 47.47%.

[[文件:Human-trivial-1.png|500px]]

=Machine performance=

Machine performance on CSLT-TRIVIAL-I.

[[文件:Machine-trivial-1.png|500px]]

Machine performance on CSLT-DISGUISE-I.

[[文件:Machine-disguise-1.png|500px]]

=Research directions=
* Speech perception.
* Forensic examination.

=Reference=

[1] Lantian Li, Yixiang Chen, Ying Shi, Zhiyuan Tang, and Dong Wang, “Deep speaker feature learning for text-independent speaker verification,”, Interspeech 2017.

[2] Lantian Li, Dong Wang, Yixiang Chen, Ying Shing, Zhiyuan Tang, http://wangd.cslt.org/public/pdf/spkfact.pdf

[3] Dong Wang,Lantian Li,Ying Shi,Yixiang Chen,Zhiyuan Tang., "Deep Factorization for Speech Signal", https://arxiv.org/abs/1706.01777

文件:Recording instruction.pdf

2017-09-26T05:32:20Z

Zhangmiao：

文件:录音指导.pdf

2017-09-26T05:15:09Z

Zhangmiao：

ASR Status Report 2017-9-18

2017-09-18T05:58:57Z

Zhangmiao：

{| class="wikitable"
!Date!!People !! Last Week !! This Week
|-
| rowspan="9"|2017.9.4

|Jiayin Cai
||
* Absent
||
*
|-

|-
|Xiaofei Kang
||
*Test and improve the IOS APP for recording audios.
*Finish the experiment to test the machine error rate,the result is in my cvss [http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?account=kangxf&step=view_request&cvssid=629 here] .
||
*Record the audios with zhangmiao using the money from wang.
|-

|-
|Miao Zhang
||
* Finish human test website
* Design recording app with Kangxf
* T-SNE analysis
||
* Absent for school class
|-

|-
|Yanqing Wang
||
* Implementation of node-pruning.
* comparison of connection-pruning and node-pruning, see [http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?account=wangyanqing&step=view_request&cvssid=634 here]
||
* continue on relationship and comparison of connection-pruning and node-pruning.
* Implementation of long-term dropout and experiments based on it.
|-

|-
|Ying Shi
||
* group-based softmax finished [http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?account=shiying&step=view_request&cvssid=627 here]
* multi-decoding for group-based softmax (in progress)
||
* mulit-decoding for group-based softmax
* PTN
* apply Lid for group-based softmax
|-

|-
|Yixiang Chen
||
* Absent
||
*
|-

|-
|Lantian Li
||
* Go on speaker segmentation tasks, see [http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?account=lilt&step=view_request&cvssid=615]
** Make some smooth tricks (Silence limits [MDR] and window-based smooth [FAR]).
** R.T. test.
* Music / Noise detection, see [http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?account=lilt&step=view_request&cvssid=624]
||
* Package the code for speaker segmentaion.
* Go on music / noise detection tasks.
|-

|-
|Zhiyuan Tang
||
* Part theoretical study of mispronunciation detection.
* Toolbook writing.
||
* Experiments on phonetic LID.
* Experiments on mispronunciation detection
|-

|}

----------------------------------------------
{| class="wikitable"
!Date!!People !! Last Week !! This Week
|-
| rowspan="9"|2017.9.4

|Jiayin Cai
||
*Got phonetic feat from a stronger phonetic network
*Finished part of the experiment using stronger phonetic feature.
||
*Will be absent for school.
*But I will finish the remaining experiment.
|-

|-
|Xiaofei Kang
||
* improve the human Test website：, save the test recordings, decline the positive samples
* Recording and cutting the audios, a total of 12 groups
||
* Continue to record the audios with zhangmiao
* Continue to ask people to do human test
|-

|-
|Miao Zhang
||
* Perform human test
* Record some other people and do the experiments again
||
* Continue to ask people to do human test
* Recording(the goal is to record 400 to 500 people) [http://cslt.riit.tsinghua.edu.cn/mediawiki/images/c/cc/录音说明.pdf here]
|-

|-
|Yanqing Wang
||
* Absent
||
*
|-

|-
|Ying Shi
||
* multi-decoding ASR model with more pdfs. Performance better than before but not well enough
* add sperate symbel to discriminated kazak and uyghur word set
* group-based softmax(in progress)
||
* finish group-based softmax and test the performance
|-

|-
|Yixiang Chen
||
* Absent
||
*
|-

|-
|Lantian Li
||
* Go on speaker segmentation tasks, see [http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?account=lilt&step=view_request&cvssid=615 here]
** Complete the phonetic-aware speaker segmentation.
*** Word-level boundaries from the ASR.
*** Word-level d-vector and clustering.
||
* Try some smooth tricks.
|-

|-
|Zhiyuan Tang
||
* Organized the code and doc of Parrot system[http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?account=tangzy&step=view_request&cvssid=635]
||
* Theoretical study of pronunciation detection
|-

|}

2017-09-11T06:38:38Z

Zhangmiao：

ASR Status Report 2017-9-11

2017-09-11T06:37:02Z

Zhangmiao：

{| class="wikitable"
!Date!!People !! Last Week !! This Week
|-
| rowspan="9"|2017.9.4

|Jiayin Cai
||
*Got phonetic feat from a stronger phonetic network
*Finished part of the experiment using stronger phonetic feature.
||
*Will be absent for school.
*But I will finish the remaining experiment.
|-

|-
|Xiaofei Kang
||
* improve the human Test website：, save the test recordings, decline the positive samples
* Recording and cutting the audios, a total of 12 groups
||
* Recording the 440 groups audios left with zhangmiao
* Continue to ask people to do human test
|-

|-
|Miao Zhang
||
* Perform human test
* Record some other people and do the experiments again
||
* Continue to ask people to do human test
* Recording(the goal is to record 400 to 500 people)[[媒体文件:范例.ogg]]
|-

|-
|Yanqing Wang
||
*
||
*
|-

|-
|Ying Shi
||
* multi-decoding ASR model with more pdfs down. Performance better than before but not well enough
* add sperate symbel to discriminated kazak and uyghur
* group-based softmax(in progress)
||
* finish group-based softmax and test the performance
|-

|-
|Yixiang Chen
||
*
||
*
|-

|-
|Lantian Li
||
*
||
*
|-

|-
|Zhiyuan Tang
||
*
||
*
*
|-

|}

----------------------------------------------
{| class="wikitable"
!Date!!People !! Last Week !! This Week
|-
| rowspan="9"|2017.9.4

|Jiayin Cai
||
*Finished the phonetic i-vector experiment.
||
*get BN feature and train i-vector LID.
*Get phonetic feat from a stronger phonetic network
*combine PTN and phonetic i-vector.
|-

|-
|Xiaofei Kang
||
* cutting audio and marking：21 speakers，a total of 1050 sentences
* Finish the new speaker recognition using the two recordings.
||
* improve the human Test website
|-

|-
|Miao Zhang
||
* Absent
||
* Perform human test on 21-style speech(add the disguise)
* Draw spectrums and t-SNE plots compared with experiment results
|-

|-
|Yanqing Wang
||
* Absent.
||
*
|-

|-
|Ying Shi
||
* multi decodeing ASR model
* multi decodeing with fake Lid [http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?account=shiying&step=view_request&cvssid=627 here]
* read code about TTS
||
* employ group softmax to train multi decoding ASR model
* synthesis one 'real' speech
|-

|-
|Yixiang Chen
||
* Absent.
||
*
|-

|-
|Lantian Li
||
* Go on speaker segmentation tasks, see [http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?account=lilt&step=view_request&cvssid=615 here]
** Dimensionality reduction.
** Clustering.
** Visualization.
||
* Phonetic-aware speaker segmentation.
|-

|-
|Zhiyuan Tang
||
* more indicators for VV scoring system, see [http://cslt.riit.tsinghua.edu.cn/mediawiki/images/a/a1/VV_scoring.pdf].
||
* more indicators, a demo with Shuai.
* toolbook writing.
|-

|}

2017-09-05T03:36:25Z

Zhangmiao：

Asr-progress 2017.08

2017-09-04T05:36:53Z

Zhangmiao：/* Time Off Table */

===Daily Report===

{|class="wikitable"
! Date !! Person !! start!! leave !! hours ||status (problems/solutions)
|-
| rowspan="5"|2017.8.1
|Yanqing Wang|| 11:00 || 19:30 || 8.5h || start to write TRP ( connection sparseness )
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 9:00 || 21:00 || 12h || Finish experiments of 12-style speech
|-
|Xiaofei Kang || 10:00 || 23:00 || 13h || discuss the recording plan and decide a preliminary plan.
|-
|Jiayin Cai || 9:00 || 22:00 || 13h || run basic PTN system
|-

|-
| rowspan="5"|2017.8.2
|Yanqing Wang || 11:00 || 19:00 || 8h || write TRP ( connection sparseness )
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 9:00 || 21:00 || 12h || Finish experiments of 12-style speech
|-
| Xiaofei Kang || 10:00 || 23:00 || 13h || Complete a part of the recording work: collecting six types of sound from 13 people.
|-
|Jiayin Cai || 9:00 || 22:00 || 13h || run basic PTN system
|-

|-
| rowspan="5"|2017.8.3
|Yanqing Wang || 11:00 || 17:00 || 6h || write TRP ( connection sparseness )
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 9:00 || 21:00 || 12h || Analyse the experiment results
|-
| Xiaofei Kang || 10:00 || 23:00 || 13h || read related paper and do a speaker recognition experiment.
|-
|Jiayin Cai || 9:00 || 22:00 || 13h || change the chunk width and run PTN
|-

|-
| rowspan="5"|2017.8.4
|Yanqing Wang || 11:00 || 19:00 || 8h || write TRP ( connection sparseness )
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 9:00 || 21:00 || 12h || Design the new test website
|-
| Xiaofei Kang || 10:00 || 23:00 || 13h || Finish experiments of 12-style speech with ZhangMiao,a total of 5 experiments.
|-
|Jiayin Cai || 9:00 || 22:00 || 13h || change the chunk width and run PTN
|-

|-
| rowspan="5"|2017.8.5
|Yanqing Wang || 12:30 || 19:00 || 6.5h || write TRP ( connection sparseness )
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 10:00 || 21:00 || 11h || Build up the website
|-
| Xiaofei Kang || || || ||
|-
|Jiayin Cai || 9:00 || 22:00 || 13h || change the chunk width and run PTN
|-

|-
| rowspan="5"|2017.8.6
|Yanqing Wang || || || ||
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || || || ||
|-
| Xiaofei Kang || || || ||
|-
|Jiayin Cai || || || ||
|-

|-
| rowspan="5"|2017.8.7
|Yanqing Wang || 11:00 || 19:00 || 8h || write a shell script to prune a new network according to a pruned network
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 9:00 || 21:00 || 12h || Record work
|-
| Xiaofei Kang || 10:00 || 23:00 || 13h || Recording the audio for the speaker recognition.
|-
|Jiayin Cai || 9:00 || 22:00 || 13h || change sample_per_iter and run PTN
|-

|-
| rowspan="5"|2017.8.8
|Yanqing Wang || 11:00 || 19:00 || 8h || use the yesterday's shell script but find that the proficiency is too low
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 9:00 || 21:00 || 12h || Wirte the code to cut silence in speech
|-
| Xiaofei Kang || 10:00 || 23:00 || 13h || continue to record the audio.
|-
|Jiayin Cai || 9:00 || 21:00 || 12h || change sample_per_iter and run PTN
|-

|-
| rowspan="5"|2017.8.9
|Yanqing Wang || 11:00 || 19:00 || 8h || learn to write a kaldi-command
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 9:00 || 21:00 || 12h || replenish the test speech
|-
| Xiaofei Kang || 10:00 || 23:00 || 13h || read the paper about i-vector and d-vector.
|-
|Jiayin Cai || 9:00 || 22:00 || 13h || run basic i-vector system
|-

|-
| rowspan="5"|2017.8.10
|Yanqing Wang || 11:00 || 19:00 || 8h || write a kaldi-command to prune a new network according to a pruned network
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 9:00 || 21:00 || 12h || replace the speech in website with new one
|-
| Xiaofei Kang || 10:00 || 23:00 || 13h || Learn the new test website from zhangmiao
|-
|Jiayin Cai || 9:00 || 22:00 || 13h || run basic i-vector system
|-

|-
| rowspan="5"|2017.8.11
|Yanqing Wang || 11:00 || 19:00 || 8h || use yesterday's command to prune a new network
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 9:00 || 21:00 || 12h || check the linear chapter
|-
| Xiaofei Kang || 10:00 || 23:00 || 13h || Learn PHP, and test the new test website.
|-
|Jiayin Cai || 9:00 || 22:00 || 13h || write the code to make phonetic feature
|-

|-
| rowspan="5"|2017.8.12
|Yanqing Wang || || || ||
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 11:00 || 20:00 || 9h || check the linear chapter
|-
| Xiaofei Kang || || || ||
|-
|Jiayin Cai || 13:00 || 22:00 || 9h || change the structure of 6 layer TDNN
|-

|-
| rowspan="5"|2017.8.13
|Yanqing Wang || || || ||
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || || || ||
|-
| Xiaofei Kang || || || ||
|-
|Jiayin Cai || 13:00 || 22:00 || 9h || run phonetic i-vector experiment for dev_all
|-

|-
| rowspan="5"|2017.8.14
|Yanqing Wang || 12:00 || 18:00 || 6h || exp: 97% pct prune and a contrast exp: apply its structure to a new network
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 9:00 || 21:00 || 12h || cut new recorded speech
|-
| Xiaofei Kang || || || ||
|-
|Jiayin Cai || 13:00 || 22:00 || 9h || run phonetic i-vector experiment for dev_1s
|-

|-
| rowspan="5"|2017.8.15
|Yanqing Wang || 12:00 || 18:00 || 6h || exp: 97% pct prune and a contrast exp: apply its structure to a new network
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 9:00 || 21:00 || 12h || run the experiments on recorded speech
|-
| Xiaofei Kang || || || ||
|-
|Jiayin Cai || 13:00 || 22:00 || 9h || write code to deal with trails.trl for test data dev_3s
|-

|-
| rowspan="5"|2017.8.16
|Yanqing Wang || 12:00 || 20:00 || 8h || summarize the 2 exp I did yesterday
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 9:00 || 21:00 || 12h || design the 20-style human test website
|-
| Xiaofei Kang || || || ||
|-
|Jiayin Cai || 13:00 || 22:00 || 9h || run phonetic i-vector experiment for dev_3s
|-

|-
| rowspan="5"|2017.8.17
|Yanqing Wang || 12:00 || 20:00 || 8h || explore the distribution of the nonlin6's output
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 9:00 || 21:00 || 12h || wirte the wbsite
|-
| Xiaofei Kang || || || ||
|-
|Jiayin Cai || 9:00 || 22:00 || 13h || run phonetic i-vector experiment for dev_3s
|-

|-
| rowspan="5"|2017.8.18
|Yanqing Wang || 12:00 || 20:00 || 8h || explore the distribution of the nonlin6's output
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 9:00 || 21:00 || 12h || wite the website
|-
| Xiaofei Kang || || || ||
|-
|Jiayin Cai || 9:00 || 22:00 || 13h || prepare the ppt for report
|-

|-
| rowspan="5"|2017.8.19
|Yanqing Wang || || || ||
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || 10:00 || 20:00 || 10h || finish the website and update the results
|-
| Xiaofei Kang || || || ||
|-
|Jiayin Cai || 10:00 || 21:00 || 11h || sort out the data for all the experiment, deal with ppt for report
|-

|-
| rowspan="5"|2017.8.20
|Yanqing Wang || || || ||
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || || || ||
|-
| Xiaofei Kang || || || ||
|-
|Jiayin Cai || 13:00 || 22:00 || 9h || deal with ppt for report, draw diagrams for report
|-

|-
| rowspan="5"|2017.8.21
|Yanqing Wang || 12:00 || 20:00 || 8h || do exps: prune a just-randomly-initiated network randomly, and train it ( groups 1 )
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || || || ||
|-
| Xiaofei Kang || 10:30 || 23:00 || 12.5h || Improve the test website to judge before committing.
|-
|Jiayin Cai || 9:00 || 22:00 || 13h || deal with ppt for report, report my experiment for my group
|-

|-
| rowspan="5"|2017.8.22
|Yanqing Wang || 11:30 || 18:00 || 6.5h || do exps: prune a just-randomly-initiated network randomly, and train it ( groups 2-3 )
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || || || ||
|-
| Xiaofei Kang || 9:30 || 21:30 || 12h || Recording new audios for the speaker recognition.
|-
|Jiayin Cai || || || ||
|-

|-
| rowspan="5"|2017.8.23
|Yanqing Wang || 11:30 || 18:00 || 6.5h || do exps: prune a just-randomly-initiated network randomly, and train it ( groups 4-5 )
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || || || ||
|-
| Xiaofei Kang|| 9:30 || 21:30 || 12h || continue to record new audios,a total of 38 person.
|-
|Jiayin Cai || || || ||
|-

|-
| rowspan="5"|2017.8.24
|Yanqing Wang || 11:30 || 18:00 || 6.5h || write awk , shell scripts and kaldi-command as a preparation for node-sparseness task.
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || || || ||
|-
| Xiaofei Kang || 9:30 || 21:30 || 12h || Organize the voice and do a sample test,and learn 4 papers from lantian
|-
|Jiayin Cai || || || ||
|-

|-
| rowspan="5"|2017.8.25
|Yanqing Wang || || || ||
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || || || ||
|-
| Xiaofei Kang || 9:30 || 21:30 || 12h || do a speaker recognition with the new audios.
|-
|Jiayin Cai || || || ||
|-

|-
| rowspan="5"|2017.8.26
|Yanqing Wang || 8:30 || 15:00 || 6.5h || survey on node-sparseness, sammarize the exps of this week and ask for short-time leave
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || || || ||
|-
| Xiaofei Kang || || || ||
|-
|Jiayin Cai || || || ||
|-

|-
| rowspan="5"|2017.8.27
|Yanqing Wang || || || ||
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || || || ||
|-
| Xiaofei Kang || || || ||
|-
|Jiayin Cai || || || ||
|-

|-
| rowspan="5"|2017.8.28
|Yanqing Wang || || || ||
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || || || ||
|-
| Xiaofei Kang || 9:30 || 21:30 || 12h || I arrange a new voice and data sets, and find out all the twice recording all of 21 people.
|-
|Jiayin Cai || || || ||
|-

|-
| rowspan="5"|2017.8.29
|Yanqing Wang || || || ||
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || || || ||
|-
| Xiaofei Kang || 9:30 || 21:30 || 12h || cutting the audio of 21 speakers, a total of 1050 sentences.
|-
|Jiayin Cai || || || ||
|-

|-
| rowspan="5"|2017.8.30
|Yanqing Wang || || || ||
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || || || ||
|-
| Xiaofei Kang || 10:30 || 23:30 || 13h || mark the segments of 1050 audios.
|-
| Jiayin Cai || 10:30 || 19:30 || 9h || do the experiment for i-vector with delta_order =3
|-

|-
| rowspan="5"|2017.8.31
|Yanqing Wang || || || ||
|-
|Yixiang Chen || || || ||
|-
|Miao Zhang || || || ||
|-
| Xiaofei Kang || 10:30 || 23:30 || 13h || finish the speaker recognition experiment,and get a test result.
|-
| Jiayin Cai || 10:30 || 19:30 || 9h || do the experiment for i-vector with delta_order =3
|-
|}

===Time Off Table===
{| class="wikitable"
! Name !! Days off
|-
|Yanqing Wang||
|-
|Yixiang Chen||
|-
|Miao Zhang||
|-
|Xiaofei Kang ||
|-
|Jiayin Cai ||
|-
|}

2017-08-05T03:19:48Z

Zhangmiao：

2017-07-17T04:18:57Z

Zhangmiao：

{| class="wikitable"
!Date!!People !! Last Week !! This Week
|-
| rowspan="7"|2017.7.17

|Miao Zhang
||
* Read the paralinguistic paper and material from Teacher Li
* work out the recording plan (delayed)
||
* work out the recording plan with instruction from Teacher Li
*check the book of deep learning
|-

|-
|Hui Tang
||
*
||
*
|-

|-
|Yanqing Wang
||
* check the former conclusions in a narrow network ( not finished yet )
* read the source code as a preparation for retrain task.
||
* finish checking the former conclusions and try to find the applicable conditions.
* finish the retrain task.
|-

|-
|Ying Shi
||
*
||
*
|-

|-
|Yixiang Chen
||
*
||
*
|-

|-
|Lantian Li
||
*
||
*
|-

|-
|Zhiyuan Tang
||
* Replace the old version kaldi with new ones. (delayed)
* Gather Part 1: 'Speech, Speech Processing and Tools' of Kaldi book for further release. (delayed)
||
* Replace the old version kaldi with new ones.
* Gather Part 1: 'Speech, Speech Processing and Tools' of Kaldi book for further release.
|-

|}

------------------------

{| class="wikitable"
!Date!!People !! Last Week !! This Week
|-
| rowspan="7"|2017.7.10

|Miao Zhang
||
* A report about trivial events
* Finish the test website with Tanghui
||
* Read the paper of Paralinguistics
* Make a plan for recording and start to record hopefully.
|-

|-
|Hui Tang
||
* completed the test web site [http://192.168.0.84:8091/speech/index.php web]
* finished checking the subset of our speech databases (nearly 800 sentences)
||
* finish checking the reminder of the databases (nearly 2500 sentences)
|-

|-
|Yanqing Wang
||
* use different activation function to prune [http://cslt.riit.tsinghua.edu.cn/mediawiki/images/0/03/Results.pdf result]
||
* make the network narrow and test the former conclusions.
* change source code to retrain the neural network.
|-

|-
|Ying Shi
||
* help zheling to finish his first crawler program
* chech hazak speech data
** train: 1346 utterances' WER large than 20% (utterance level WER)
** test: 759 utterances' WER large than 20%(utterance level WER)
* transfer learning based on th30 and wsj (performance is poor)
||
* tools for speech data checking
* transfer learning based on large Chinese ASR model
|-

|-
|Yixiang Chen
||
* plot for “learning deep speaker features”
||
* comprehend Paralinguistics
* record voice
|-

|-
|Lantian Li
||
* deep speaker feature
** segmentation is still not suitable.
** visualization with the t-sne seems cool.
* help Zhangzy decode d-vector and re-train a new deep speaker model.
||
* more details of segmentation experiments.
* prepare the weekly meeting.
|-

|-
|Zhiyuan Tang
||
* Scanned the source code of auto-scoring system;
* A report about the research of the speech group (Thursday).
||
* Replace the old version kaldi with new ones.
* Gather Part 1: 'Speech, Speech Processing and Tools' of Kaldi book for further release.
|-

|}