2013-10-25 - 版本历史

Cslt：/* Noisy training */

2013-10-25T01:48:53Z

‎Noisy training

2013年10月25日 (五) 01:47 Cslt

2013-10-25T01:47:59Z

2013年10月25日 (五) 01:47 Cslt

2013-10-25T01:47:25Z

2013年10月25日 (五) 01:46 Cslt

2013-10-25T01:46:59Z

Cslt：以内容“== Data sharing == * LM count files still undelivered! == DNN progress == === Sparse DNN === * Optimal Brain Damage(OBD). # The initial observation shows that, af...”创建新页面

2013-10-25T01:45:53Z

以内容“== Data sharing == * LM count files still undelivered! == DNN progress == === Sparse DNN === * Optimal Brain Damage(OBD). # The initial observation shows that, af...”创建新页面

新页面

== Data sharing ==

* LM count files still undelivered!

== DNN progress ==

=== Sparse DNN ===

* Optimal Brain Damage(OBD).

# The initial observation shows that, after direct/obd cutting, obd is better than direct cutting in terms of training and testing accuracy.
# The sticky training goes.

=== Tencent exps ===
N/A

==Noisy training ==

# The 863 clean test shows consistent performance.
# Investigation shows that the car noise corrupted the speech patterns more seriously with the current fb-based noise adding. This leads to worse performance with car noises than with white noises. Note this observation is valid only with the fb-based noise adding

==Continuous LM ==

# The lattice rescore toolkit is done. The efficiency problem is solved by DP.
# The perplexity of 10-NN is worse than 1-NN
# Initial lattices rescoring shows that the rescoring needs remove the old ngram score and then insert the new score. The 1st nn works better than the 10-nn.
# The CSLM does not improve the performance so far. This is conflict to the ppl results.
# Investigate better ways to combine multiple NNs. Investigate rescoring n-best results instead of lattices.

<pre>
lattice-rescore: traing-data:500M,dict:11w,test:tencent

id cslm(lopp) cslm(-logp) n-gram n-gram(replace) cslm(-old+cslm) cslm((0-1024)) cslm((0-1024),(1024-2048))
map: 53.86 38.44 34.40 37.38 35.38 34.94 34.91
2044: 47.81 32.20 27.18 31.18 28.53 27.65 27.87
notetp3: 43.01 24.61 18.78 22.34 20.18 19.64 19.81
record1900: 32.14 20.42 12.89 20.10 13.97 13.52 13.81
general: 62.86 49.17 43.23 47.50 45.55 44.28 44.62
online1: 62.79 46.76 39.49 46.40 40.26 39.91 40.01
online2: 58.20 39.21 32.15 38.75 33.04 32.61 32.65
speedup: 52.50 38.74 31.47 37.72 34.01 32.65 32.81

note: cslm(logp) --replace lattice value1 with cslm lnp
cslm(-logp)--replace lattice value1 with cslm -lnp
ngram(replace)--replace lattice vaule1 with ngram -lnp
cslm((-old+cslm)--repace lattice value1 = -old_ngram+cslm
cslm((0-1024))--replace lattice value1=-old_ngram+cslm (slist:0-1024,mach_num=1)
cslm((0-1024),(1024-2048))--replace lattice value1=-old_ngram+cslm (slist:0-1024,1024-1048,mach_num=2)

</pre>
==QA LM==

# Tencent word segmentation system ready.
# Collecting data for Q-LM training.

@@ 第22行： / 第22行： @@
 ==LM development==
-===Continuous LM ===
 # The lattice rescore toolkit is done. The efficiency problem is solved by DP.
@@ 第52行： / 第53行： @@
 </pre>
 ===QA LM===
 #  Tencent word segmentation system ready.
 #  Collecting data for Q-LM training.

@@ 第15行： / 第15行： @@
 # The 863 clean test shows consistent performance.
-# Investigation shows that the car noise corrupted the speech patterns more seriously with the current fb-based noise adding. This leads to worse performance with car noises than with white noises. Note this observation is valid only with the fb-based noise adding
+# Investigation shows that the car noise corrupted the speech patterns more seriously with the current fb-based noise adding. This leads to worse performance with car noises than with white noises. Note this observation is valid only with the fb-based noise adding.
 === Tencent exps ===

@@ 第3行： / 第3行： @@
 * LM count files still undelivered!
-== DNN progress ==
+== AM development ==
 === Sparse DNN ===