Monaural Speech Enhancement Using Intra-Spectral Recurrent Layers in the Magnitude and Phase Responses

Nayem, Khandokar Md.; Williamson, Donald S.

doi:10.1109/icassp40776.2020.9054298

Cited by 2 publications

(1 citation statement)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, although recurrent networks capture temporal correlations, these approaches do not ensure that the resulting speech spectra exhibit realistic spectral-and temporal-fine structure that occurs within real speech signals from human sources. Our recent work strives to enforce spectral-fine structure by incorporating an intra-spectral recurrence layer [14,15], but they do not address temporal-fine structure and do not ensure that humanlike spectra is generated.…”

Section: Introductionmentioning

confidence: 99%

Towards An ASR Approach Using Acoustic and Language Models for Speech Enhancement

Nayem

Williamson

2021

ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

Self Cite

View full text Add to dashboard Cite

Recent work has shown that deep-learning based speech enhancement performs best when a time-frequency mask is estimated. Unlike speech, these masks have a small range of values that better facilitate regression-based learning. The question remains whether neural-network based speech estimation should be treated as a regression problem. In this work, we propose to modify the speech estimation process, by treating speech enhancement as a classification problem in an ASR-style manner. More specifically, we propose a quantized speech prediction model that classifies speech spectra into a corresponding quantized class. We then train and apply a language-style model that learns the transition probabilities of the quantized classes to ensure more realistic speech spectra. We compare our approach against time-frequency masking approaches, and the results show that our quantized spectra approach leads to improvements.

show abstract

Section: Introductionmentioning

confidence: 99%