阅读"voice conversion using deep bidirectional long short-term memory based recurrent neural networks"

1. VC的两种方法：
   
   • 基于规则：修改语音信号
     
     • ”Frequency warping based on mapping formant parameters”
     • “Weighted frequency warping for voice conversion”
   • 基于统计：估计从源信号到目的信号的map函数
     
     • GMM：原理参考基于高斯混合模型的语音转换技术研究
       
       • ”Continuous probablistic transform for voice conversion”
       • “Voice conversion based on maximum-likelihood estimation of spectral parameter trajectory” 使用动态特征和全局方差
     • DBN：
       
       • ”A fast learning algorithm for Deep Belief Nets”
       • “Acoustic modeling using Deep Belief Networks”
       • “Voice conversion in high-order eigen space using Deep Belief Nets”
     • RMB：
       
       • ”Joint spectral distribution modeling using Restricted Boltzmann Machines for voice conversion”
       • “Voice conversion using Deep Neural Networks with layer-wise
         generative training”
     • RNN：
       
       • High-order sequence modeling using speaker-dependent recurrent temporal Restricted Boltsmann Machines for voice conversion
       • 缺点：只能利用previous context而不是future context, 切由于vanishing and exploding gradients(参考”Learning long-
         term dependencies with gradient descent is difficult”)，也不能处理长序列
     • BLSTM-RNN：Bidirectional Long Short-Term Memory
       
       • ”Framewise phoneme classification with bidirectional LSTM and other neural network architectures”
       • “Long Short-Term Memory”
2. 传统RNN模型
3. Bidirectional RNN模型：利用sequence
4. LSTM模型
5. BRNN与LSTM结合
6. 6.