Noise Robust Automatic Speech Recognition Based on Spectro-Temporal Techniques

Kovács, George L.

doi:10.14232/phd.4108

Cited by 1 publication

(2 citation statements)

References 142 publications

(224 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This is followed by a short discussion of the merger network itself. It should be noted that these discussions are kept intentionally short, as detailed description of these methods are available in the earlier publications of the authors (see [18,19]) Lastly, we finish the section by a detailed description of the band selection methods applied for choosing the bands to be dropped.…”

Section: Methodsmentioning

confidence: 99%

“…Motivated by psycho-acoustic evidence [16], signal processing considerations [17], and the potential for parallelization, in the multi-band processing paradigm (see Figure 1) the input speech signal is first decomposed into spectral bands, then these bands are processed independently before the information from different bands is merged in order to produce the aimed recognition result. There are thus three key issues to be addressed in this paradigm, namely the approach used for separating the speech signal into bands, the method of band processing, and lastly, the mechanism used for combining the information extracted from the different bands [18]. In this study, we follow the pipeline described in Kovács et al [19].…”

Section: Multi-band Speech Recognitionmentioning

confidence: 99%

See 1 more Smart Citation

Examining the Combination of Multi-Band Processing and Channel Dropout for Robust Speech Recognition

Kovács¹,

Tóth²,

Compernolle³

et al. 2019

Interspeech 2019

Self Cite

View full text Add to dashboard Cite

A pivotal question in Automatic Speech Recognition (ASR) is the robustness of the trained models. In this study, we investigate the combination of two methods commonly applied to increase the robustness of ASR systems. On the one hand, inspired by auditory experiments and signal processing considerations, multi-band band processing has been used for decades to improve the noise robustness of speech recognition. On the other hand, dropout is a commonly used regularization technique to prevent overfitting by keeping the model from becoming over-reliant on a small set of neurons. We hypothesize that the careful combination of the two approaches would lead to increased robustness, by preventing the resulting model from over-rely on any given band. To verify our hypothesis, we investigate various approaches for the combination of the two methods using the Aurora-4 corpus. The results obtained corroborate our initial assumption, and show that the proper combination of the two techniques leads to increased robustness, and to significantly lower word error rates (WERs). Furthermore, we find that the accuracy scores attained here compare favourably to those reported recently on the clean training scenario of the Aurora-4 corpus.

show abstract

Section: Methodsmentioning

confidence: 99%