FreeNeb Status Report 2016-12-12

From cslt Wiki
Revision as of 07:21, 12 December 2016 by Cslt (Talk | contribs)

Jump to: navigation, search

This week

People Last Week This Week
Chao Xing
  • TTS
  • TTS
Ziwei Bai
Mengyuan Zhao
  • Compile kaldi for ARM11 board and delivered to roobo with a grammar-based decoding graph and an initial version of AM.
  • Form our regulation with Dr.Wang, ZZY, and prepare on-board package for new member.
  • Buy and install a PC for our new member Shugen Wang.
  • Supervise tonghuashun's 2000h AM training.
  • Investigate chain model.
Zhiyong Zhang
Yang Wei
  • Rewrite cleaning script to improve the performance, and use it to train new LM. The test result were improved a little.
  • Modify tdnn and tdnn_discriminative script to solve problem.
  • Modify tdnn_discriminative script to solve problem.
  • Run tdnn script from chain
Dong Wang
  • Deal with Roobo project. Contract almost done.
  • To Shenzhen, for the second term collaboration. Looks good.
  • Suggest for TTS. Seems an initial version with rich context is fine.
  • No show. Will be on APSIPA.
  • May need to deal with the Tonghuashun progress stuff.
Shugen Wang
  • Familiar with fst.
  • Study materials of deep learning.
  • Select the useful files in fst library.

Last week

People Last Week This Week
Chao Xing
  • TTS
  • TTS dnn method
Ziwei Bai
  • modify the CDSSM code
  • test the CDSSM model
Mengyuan Zhao
  • Finish svd and MPE training of pdf-3318 AM.
  • Investigate kaldi, openfst, openblas cross-compile on ARM11 board.
  • Continue investigate kaldi cross-compile.
  • Draft welcome letter for new colleague, and develop entry procedures.
  • Train 2000h AM for Tonghuashun (if data arrive).
Zhiyong Zhang
  • Checked and modified part of ver0.1 vivi-decoder specific document;
  • Finished the stream module of vivi-decoder.
  • To finished the vivi-decoder specific document of ver2;
  • Continue to do some investigation on embedding decoder.
Yang Wei
  • Adjust the tdnn script from chain to train AM
  • Continue training AM with chain
  • Rewrite cleaning script in order to improve performance of the LM.
Dong Wang
  • Tonghuashun data delivery negotiation
  • Roboo project negotiation
  • Smart station project discussion
  • Roboo project strategy discussion