2013-12-13 - 版本历史

2013年12月16日 (一) 09:54 Cslt

2013-12-16T09:54:51Z

2013年12月13日 (五) 08:12 Cslt

2013-12-13T08:12:53Z

Cslt：/* Speech QA */

2013-12-13T01:35:30Z

‎Speech QA

2013年12月13日 (五) 01:34 Cslt

2013-12-13T01:34:42Z

Cslt：以内容“== AM development == === Sparse DNN === * Optimal Brain Damage(OBD). # Online OBD held. # OBD + L1 norm start to investigation. * Efficient computing # Using MK...”创建新页面

2013-12-13T01:33:07Z

以内容“== AM development == === Sparse DNN === * Optimal Brain Damage(OBD). # Online OBD held. # OBD + L1 norm start to investigation. * Efficient computing # Using MK...”创建新页面

新页面

== AM development ==

=== Sparse DNN ===

* Optimal Brain Damage(OBD).

# Online OBD held.
# OBD + L1 norm start to investigation.

* Efficient computing

# Using MKL and CSR storage does not help much for sparse matrix computation. When the sparsity is 20%, the computing costs 2 times of the original time.
# Using matrix splitting can improve computing performance for sparse matrix. Using BSR (block sparse row), when the sparsity is 1/6, the same time cost was obtained.
# We can re-arrange the matrix structure and compose zero blocks by some smart approaches, leading to better computing speed.
# There is minor difference between the MKL computing and direct computing. This means computing accuracy does not impact the ASR performance very much. This give some excuse for extremely sparse matrix construction.

=== Efficient DNN training ==

# Moment-based training. NN accuracy decreased with larger moment, but ASR performance increased.
# Asymmetic window: left 20, right 5. NN accuracy increase by 7%.

===Engine optimization===

* Investigating LOUDS FST.

==LM development==

===NN LM===

* 3 iteration 500 M training done. 24 hours per iteration.
* PPL 189 after 3 iterations.
* NN-based CSLM merge done (10240*100*10240). The PPL and WER are both worse than the original 10 network outputs.
* Need to investigate why the merge is not accurate.

==Embedded development==

* Embedded stream mode on progress.

==Speech QA==

* Text-based QA. 121/199 correction (with answers). 58 no answers(24 no attributes in db, 27 no recorder). 20 with incorrect answers(5 no answers in the db and obtain incorrect from the web, 8 no recorder and obtain incorrect from the web, 3 db error).
* Speech-based QA. WER=8.70%. SEE=32.0%. Almost English wrong. Remove English SEE=27.1%.
* Use all the correct input. 45.14% correct in all the input (18*199)
* Will try to recover some ASR errors using QA.

@@ 第13行： / 第13行： @@
 # Using matrix splitting can improve computing performance for sparse matrix. Using BSR (block sparse row), when the sparsity is 1/6, the same time cost was obtained.
 # We can re-arrange the matrix structure and compose zero blocks by some smart approaches, leading to better computing speed.
-# There is minor difference between the MKL computing and direct computing. This means computing accuracy does not impact the ASR performance very much. This give some excuse for extremely sparse matrix construction.
+# There is minor difference between the MKL computing and direct computing. This means computing accuracy does not impact the ASR performance very much. This give some rationality for extremely sparse matrix construction.
 === Efficient DNN training ===
-# Moment-based training. NN accuracy decreased with larger moment, but ASR performance increased.
+# Moment-based training. NN accuracy decreased with a larger moment, but ASR performance increased (e.g., 0.2).
-# Asymmetic window: left 20, right 5. NN accuracy increase by 7%.
+# Asymmetric window: left 20, right 5. NN accuracy increase by 7%.
@@ 第30行： / 第30行： @@
 ===NN LM===
-* bigger CSLM with 10240 words output. Performance is better than the separate trained 10 networks (and merge).
+* bigger CSLM with 10240 words output. Performance is better than the separately trained 10 networks (and merge).
@@ 第40行： / 第40行： @@
 ==Speech QA==
-* Clustered based QA LM using data from Q db.
+* Class-based QA LM using data from Q db is done.
-* Extract some documents from Baidu known related to music.
+* Extract some documents from Baidu know-how that are related to music.
-* Text-based QA. 121/199 correction (with answers). 58 no answers(24 no attributes in db, 27 no recorder). 20 with incorrect answers(5 no answers in the db and obtain incorrect from the web, 8 no recorder and obtain incorrect from the web, 3 db error).
+* Text-based QA. 121/199 correction (with answers). 58 no answers(24 no attributes in db, 27 no recorders). 20 with incorrect answers(5 no answers in the db and so obtained incorrect answers from the web; 8 no recorder and so obtained incorrect answers from the web; 3 db error).
-* Speech-based QA. WER=8.70%. SEE=32.0%. Almost English wrong. Remove English SEE=27.1%.
+* Speech-based QA. WER=8.70%. SEE=32.0%. Almost English queries are wrong. Remove English, SEE=27.1%.
-* SP-QA accuracy 45.14% in all the input (18*199)
+* SP-QA accuracy 45.14% in all the input (18*199).
-* Will try to recover some ASR errors using QA.
+* Will try to recover some ASR errors using QA, e.g., pronunciation correction.