随笔分类 -  语音识别

摘要:1. FSMN参数规模分析 & 160; & 160; & 160; & 160; (1)分析前提 : 1. 假设隐藏层单元规模都为 n 2. 只分析前向 t 个时刻的结构,即暂时不考虑双向的结构 3. 只分析向量系数编码,即vFSMN,暂时不考虑sFSMN & 160; & 160; & 160; 阅读全文
posted @ 2019-04-08 17:50 JeffreyLee 阅读(846) 评论(0) 推荐(0)
摘要:参考文献如下: (1) A Context Sensitive Chunk BPTT Approach to Training Deep LSTM/BLSTM Recurrent Neural Networks for Offline Handwriting Recognition (2) Trai 阅读全文
posted @ 2019-03-26 15:14 JeffreyLee 阅读(1626) 评论(0) 推荐(0)
摘要:参考文献如下: (1) Deep Feed Forward Sequential Memory Networks for Speech Synthesis (2) Deep FSMN for Large Vocabulary Continuous Speech Recognition 1. cFSM 阅读全文
posted @ 2019-03-21 18:55 JeffreyLee 阅读(5478) 评论(0) 推荐(0)
摘要:参考文献如下: (1) Feedforward Sequential Memory Neural Networks without Recurrent Feedback (2) Feedforward Sequential Memory Networks: A New Structure to Le 阅读全文
posted @ 2019-03-21 17:11 JeffreyLee 阅读(5620) 评论(2) 推荐(0)