Skip to main content
EURASIP Journal on Audio, Speech, and Music Processing
Figure 4 | EURASIP Journal on Audio, Speech, and Music Processing

Figure 4

From: Deep neural network-based bottleneck feature and denoising autoencoder-based dereverberation for distant-talking speaker identification

Figure 4

Topology of denoising autoencoder for cepstral domain dereverberation. In the training phase of the DAE, the cepstrum of reverberant speech was set as the input, and the cepstrum of the corresponding clean speech was set as the teacher signal. After training, the DAE network was able to generate the output of reverberation suppressed features.

Back to article page