“ASR Status Report 2016-09-19”版本间的差异

• LSTM cell analysis[1]

• analysis of gates, layers, temporal property

• Write book

• Continue with the book writing

• nn visualization

1. bi-monthly report
2. memory length of lstm and gru
3. some work about speaker gender

• Continue research on lstm and gru(multilayer)

• Deep speaker embedding ( Random sample training RST ) [2].
• Local training [3].
• Joint training on SRE and LRE [4].

• Continue these tasks.

• memory analysis of gated recurrent units (memory term, memory residual), outline for icassp 2017

• paper for icassp 2017

--

--

• Deep speaker embedding ( Random sample training RST ) [5].
• Local training [6].
• Joint training on SRE and LRE [7].

• Paper for ICASSP 2017.

• paper for icassp17

• technical report for icassp17 (visualization analysis for recurrent networks)
• interspeech16 paper reading

--

--

• Deep speaker embedding ( Random sample training RST ) [8].
• Local training [9].
• Joint training on SRE and LRE [10].

• Paper for ICASSP 2017.