Title :
Improving DNN speaker independence with I-vector inputs
Author :
Senior, Alan ; Lopez-Moreno, Ignacio
Author_Institution :
Google Inc., New York, NY, USA
Abstract :
We propose providing additional utterance-level features as inputs to a deep neural network (DNN) to facilitate speaker, channel and background normalization. Modifications of the basic algorithm are developed which result in significant reductions in word error rates (WERs). The algorithms are shown to combine well with speaker adaptation by backpropagation, resulting in a 9% relative WER reduction. We address implementation of the algorithm for a streaming task.
Keywords :
backpropagation; feature extraction; neural nets; speech processing; vectors; DNN speaker independence; I-vector inputs; WER; background normalization; backpropagation; channel normalization; deep neural network; speaker normalization; streaming task; utterance-level features; word error rates; Adaptation models; Computational modeling; Data models; Hidden Markov models; Neural networks; Speech; Training; Deep neural networks; Voice Search; i-vectors; large vocabulary speech recognition; speaker adaptation;
Conference_Titel :
Acoustics, Speech and Signal Processing (ICASSP), 2014 IEEE International Conference on
Conference_Location :
Florence
DOI :
10.1109/ICASSP.2014.6853591
