Difference between revisions of "140428-Shi Yin"

From cslt Wiki
Jump to: navigation, search
 
Line 1: Line 1:
 
=== Accomplished this week ===
 
=== Accomplished this week ===
* Found the noisy-training directory and listed the current process.
+
* Decoding on evil92 data set on iteration 1-4.
* Partially got the models by narrow-band noisy-training.
+
* Completed the whole job about smoothing the results of VAD with DNN.
* Perfectly read the code of VAD.  
+
* Tested the results of VAD with DNN and analyzed their performance.  
 +
* Partially completed the job about integrating the code of VAD with DNN into the framework of VAD.
 +
* Read the papers about MAP and MLLR.
  
 
=== Next Week ===
 
=== Next Week ===
 
* Go on completing the job about integrating the code of VAD with DNN into the framework of standard VAD.
 
* Go on completing the job about integrating the code of VAD with DNN into the framework of standard VAD.
 
* Decoding on dev93 data set on iteration 1-4 and analyze the results.
 
* Decoding on dev93 data set on iteration 1-4 and analyze the results.
* Decoding on the models of narrow-band noisy-training.
+
* Try to fix the DNN model by MAP or MLLR.
* Try to fix the DNN model by MAP or MLLR for far-field ASR.
+
* Find the noisy-training directory and list the current process.
 +
* Go on completing the job about narrow-band noisy-training.

Latest revision as of 12:53, 5 May 2014

Accomplished this week

  • Decoding on evil92 data set on iteration 1-4.
  • Completed the whole job about smoothing the results of VAD with DNN.
  • Tested the results of VAD with DNN and analyzed their performance.
  • Partially completed the job about integrating the code of VAD with DNN into the framework of VAD.
  • Read the papers about MAP and MLLR.

Next Week

  • Go on completing the job about integrating the code of VAD with DNN into the framework of standard VAD.
  • Decoding on dev93 data set on iteration 1-4 and analyze the results.
  • Try to fix the DNN model by MAP or MLLR.
  • Find the noisy-training directory and list the current process.
  • Go on completing the job about narrow-band noisy-training.