Loading [a11y]/accessibility-menu.js
Time-frequency convolutional networks for robust speech recognition | IEEE Conference Publication | IEEE Xplore

Time-frequency convolutional networks for robust speech recognition


Abstract:

Convolutional deep neural networks (CDNNs) have consistently shown more robustness to noise and background contamination than traditional deep neural networks (DNNs). For...Show More

Abstract:

Convolutional deep neural networks (CDNNs) have consistently shown more robustness to noise and background contamination than traditional deep neural networks (DNNs). For speech recognition, CDNNs apply their convolution filters across frequency, which helps to remove cross-spectral distortions and, to some extent, speaker-level variability stemming from vocal tract length differences. Convolution across time has not been considered with much enthusiasm within the speech technology community. This work presents a modified CDNN architecture that we call the time-frequency convolutional network (TFCNN), in which two parallel layers of convolution are performed on the input feature space: convolution across time and frequency, each using a different pooling layer. The feature maps obtained from the convolution layers are then combined and fed to a fully connected DNN. Our experimental analysis on noise-, channel-, and reverberation-corrupted databases shows that TFCNNs demonstrate reduced speech recognition error rates compared to CDNNs whether using baseline mel-filterbank features or noise-robust acoustic features.
Date of Conference: 13-17 December 2015
Date Added to IEEE Xplore: 11 February 2016
ISBN Information:
Conference Location: Scottsdale, AZ, USA
References is not available for this document.

1. Introduction

Deep learning techniques [1] are now integral to current automatic speech recognition (ASR) systems [2]. Deep learning has been used for feature representation [3], acoustic modeling [1], and language modeling [4]. Although the results from deep neural networks (DNNs) have always been encouraging, current research is focused on both improving the state-of-the-art and increasing scientific understanding of deep learning's strengths and weaknesses. Although DNNs have been observed to work highly reliably under matched conditions, they are susceptible to performance degradations under mismatched conditions [28]. Speech-signal degradations (such as reverberation, noise, and channel mismatch) can significantly reduce DNN recognition accuracy, revealing DNN's vulnerability [4], [5] to unseen conditions.

Getting results...

References

References is not available for this document.