Global SNR Estimation of Speech Signals Using Entropy and Uncertainty Estimates from Dropout Networks

Rohith Aralikatti, Dilip Kumar Margam, Tanay Sharma, Abhinav Thanda and Shankar Venkatesan

Abstract:

This paper demonstrates two novel methods to estimate the global SNR of speech signals. In both methods, Deep Neural Network-Hidden Markov Model (DNN-HMM) acoustic model used in speech recognition systems is leveraged for the additional task of SNR estimation. In the first method, SNR is estimated using the entropy of the posterior distribution obtained from DNN of an ASR system. Recent work on bayesian deep learning has shown that a DNN-HMM trained with dropout can be used to estimate model uncertainty by approximating it as a deep Gaussian process. In the second method, this approximation is used to obtain model uncertainty estimates. Noise specific regressors are used to predict the SNR from the entropy and model uncertainty. The DNN-HMM is trained on GRID corpus and tested on different noise profiles from the DEMAND noise database at SNR levels ranging from -10 dB to 30 dB.

---

Cite as: Aralikatti, R., Margam, D.K., Sharma, T., Thanda, A., Venkatesan, S. (2018) Global SNR Estimation of Speech Signals Using Entropy and Uncertainty Estimates from Dropout Networks. Proc. Interspeech 2018, 1878-1882, DOI: 10.21437/Interspeech.2018-1884.

---

BiBTeX Entry:

@inproceedings{Aralikatti2018,
author={Rohith Aralikatti and Dilip Kumar Margam and Tanay Sharma and Abhinav Thanda and Shankar Venkatesan},
title={Global SNR Estimation of Speech Signals Using Entropy and Uncertainty Estimates from Dropout Networks},
year=2018,
booktitle={Proc. Interspeech 2018},
pages={1878--1882},
doi={10.21437/Interspeech.2018-1884},
url={http://dx.doi.org/10.21437/Interspeech.2018-1884} }