Sinovoice-2014-01-20 - 版本历史

2014年1月20日 (一) 10:12 Cslt

2014-01-20T10:12:02Z

2014年1月20日 (一) 08:13 Cslt

2014-01-20T08:13:38Z

2014年1月20日 (一) 07:43 Cslt

2014-01-20T07:43:11Z

Cslt：以内容“=Project management= * Xiaoming and Xiao Na were added into the mail list * Potential Huawei conference-transcribing project was discussed =DNN training= ==Environme...”创建新页面

2014-01-20T06:36:24Z

以内容“=Project management= * Xiaoming and Xiao Na were added into the mail list * Potential Huawei conference-transcribing project was discussed =DNN training= ==Environme...”创建新页面

新页面

=Project management=

* Xiaoming and Xiao Na were added into the mail list
* Potential Huawei conference-transcribing project was discussed

=DNN training=

==Environment setting==

* New disk space (3T) was created and mounted at /nfs/disk1
* Jobs with 100 threads work fine on the cluster

==Corpora==
* 60 hour data were cut this week
* Just send out to vendors for labeling
* Waiting for out-source platform construction
* We assume 60 hour data per week in future

==470 hour 8k training==

* CE training done
* MPE training partially done

{| class="wikitable"
! Model !! CE !! MPE1!! MPE2 !! MPE3 !! MPE4
|-
|4k states||23.27/22.85 || 21.35/18.87 || 21.18/18.76 || 21.07/18.54
|-
|8k states ||22.16/22.22 || - ||20.36/17.94 || - ||
|-
|}

==6000 hour 16k training==

* Audio files ready. Files with incorrect sampling rates were removed
* Lexicon and LM were ready
* Making MFCC features
* Initial model (6 iterations etc) can be delivered before the spring holiday

=DNN Decoder=
* Initial trail of DNN decoder based on the Sinovoice code was failed, largely due to FST compiler
* Change the strategy to an integrated approach: use the sinovoice system to control connections, and use Kaldi base for asr engine
* Xiaoming will do some investigation on the Sinovoice FST compiler, while Liu Chao will focus on the Kaldi-based decoder

@@ 第3行： / 第3行： @@
 ==Environment setting==
-* Cluster accounts rearrangement
+* Accounts re-arrangement done on the SGE cluster. NO ROOT TO WORK.
-* Withdraw root/sudo previelege
+* Changed NFS server to 40 processes, hope to increase disk reading.
-* Changed NFS server to 40 processes, hope to increase the disk reading speed
+* Agree to withdraw root/sudo privilege.
-* Create a RAID-0 with 3 or 4 3T disks
+* Agree to create a RAID-0 with another 3 3T disks
 ==Corpora==
-* Change the data labeling strategy: do not label gender and the length of noise in the rest of the corpora.
+* Changed the data labeling strategy: gender and noise length will not be labelled for the following several corpora.
 * Automatic labeling
-:* Xiaoming will work with Zhiyong to discover how to generate transcriptions with confidence score embedded.
+:* Xiaoming will work with Zhiyong to discover how to generate transcriptions with confidence score held.
-:* The first step is to investigate the raw accuracy on the domain-dependent test, and then decide the quality of automatic labeling
+:* The first step is to investigate the raw accuracy on the domain-dependent test, and then decide if it is appropriate to use automatic labeling
-* Xiao Na prepare 300h telephone data (Sinovoice recording) to improve the 8k model.
+* Xiao Na will prepare 300h telephone speech data (Sinovoice recording). This will be used to improve the 8k model.
@@ 第31行： / 第31行： @@
 ==6000 hour 16k training==
-* Feature extraction done: solved three problems in the data: (1) short wave (2) unmatched file length (3) unmatched sample rate
+* Feature extraction done: solved several problems in the data: (1) short wave (2) unmatched file length (3) unmatched sample rate.
-* Training goes to tri4b, quick increase of states/pdfs
+* Training has gone to tri4b, quick increase of states/pdfs.
-* DNN training could be started from Tuesday
+* DNN training will be started on Tuesday.
 =DNN Decoder=
-* Sinovoice decoder: some errors in FST building. Many triphones are lost after graph building. Problems in cdgen?
+* Sinovoice decoder: some errors in FST building. Many triphones were lost after C composing. Problems in cdgen?
 * Kaldi decoder:
-:* A minor difference between CLG/HCLG results was find. Debugging into the problem.
+:* A minor difference between CLG/HCLG results was found. Debugging into the problem.
-:* CLG RT is comparable to the HCLG RT, 0.3-0.4 in CSLT grid-2.
+:* CLG RT is comparable to the HCLG, roughly 0.3-0.4 in CSLT grid-2.
 :* Additional optimization on pdf-pre-computing will be investigated.
 :* Code deliver today.

@@ 第13行： / 第13行： @@
 :* Xiaoming will work with Zhiyong to discover how to generate transcriptions with confidence score embedded.
 :* The first step is to investigate the raw accuracy on the domain-dependent test, and then decide the quality of automatic labeling
 ==470 hour 8k training==

@@ 第1行： / 第1行： @@
 =DNN training=
 ==Environment setting==
-* New disk space (3T) was created and mounted at /nfs/disk1
+* Cluster accounts rearrangement
-* Jobs with 100 threads work fine on the cluster
+* Withdraw root/sudo previelege
 ==Corpora==
-* 60 hour data were cut this week
+* Change the data labeling strategy: do not label gender and the length of noise in the rest of the corpora.
-* Just send out to vendors for labeling
+* Automatic labeling
-* Waiting for out-source platform construction
+:* Xiaoming will work with Zhiyong to discover how to generate transcriptions with confidence score embedded.
-* We assume 60 hour data per week in future
+:* The first step is to investigate the raw accuracy on the domain-dependent test, and then decide the quality of automatic labeling
 ==470 hour 8k training==
-* CE training done
+* MPE training done
-* MPE training partially done
 {| class="wikitable"
 ! Model !! CE !! MPE1!! MPE2 !! MPE3 !! MPE4
 |-
-|4k states||23.27/22.85 || 21.35/18.87 || 21.18/18.76 || 21.07/18.54
+|4k states ||23.27/22.85 || 21.35/18.87 || 21.18/18.76 || 21.07/18.54 || 20.93/18.32
 |-
-|8k states ||22.16/22.22 || - ||20.36/17.94 || - ||
+|8k states ||22.16/22.22 || 20.55/18.03 ||20.36/17.94  || 20.32/17.78 || 20.29/17.80
 |-
 |}
@@ 第33行： / 第29行： @@
 ==6000 hour 16k training==
-* Audio files ready. Files with incorrect sampling rates were removed
+* Feature extraction done: solved three problems in the data: (1) short wave (2) unmatched file length (3) unmatched sample rate
-* Lexicon and LM were ready
+* Training goes to tri4b, quick increase of states/pdfs
-* Making MFCC features
+* DNN training could be started from Tuesday
-* Initial model (6 iterations etc) can be delivered before the spring holiday
 =DNN Decoder=
-* Initial trail of DNN decoder based on the Sinovoice code was failed, largely due to FST compiler
-* Change the strategy to an integrated approach: use the sinovoice system to control connections, and use Kaldi base for asr engine
+* Sinovoice decoder: some errors in FST building. Many triphones are lost after graph building. Problems in cdgen?
-* Xiaoming will do some investigation on the Sinovoice FST compiler, while Liu Chao will focus on the Kaldi-based decoder
+* Kaldi decoder: