@inproceedings{dai-etal-2022-exploiting,
title = "Exploiting the compressed spectral loss for the learning of the {DEMUCS} speech enhancement network",
author = "Dai, Chi-En and
Hong, Qi-Wei and
Hung, Jeih-Weih",
editor = "Chang, Yung-Chun and
Huang, Yi-Chin",
booktitle = "Proceedings of the 34th Conference on Computational Linguistics and Speech Processing (ROCLING 2022)",
month = nov,
year = "2022",
address = "Taipei, Taiwan",
publisher = "The Association for Computational Linguistics and Chinese Language Processing (ACLCLP)",
url = "https://aclanthology.org/2022.rocling-1.13",
pages = "100--106",
abstract = "This study aims to improve a highly effective speech enhancement technique, DEMUCS, by revising the respective loss function in learning. DEMUCS, developed by Facebook Team, is built on the Wave-UNet and consists of convolutional layer encoding and decoding blocks with an LSTM layer in between. Although DEMUCS processes the input speech utterance purely in the time (wave) domain, the applied loss function consists of wave-domain L1 distance and multi-scale shorttime-Fourier-transform (STFT) loss. That is, both time- and frequency-domain features are taken into consideration in the learning of DEMUCS. In this study, we present revising the STFT loss in DEMUCS by employing the compressed magnitude spectrogram. The compression is done by either the power-law operation with a positive exponent less than one, or the logarithmic operation. We evaluate the presented novel framework on the VoiceBank-DEMAND database and task. The preliminary experimental results suggest that DEMUCS containing the power-law compressed magnitude spectral loss outperforms the original DEMUCS by providing the test utterances with higher objective quality and intelligibility scores (PESQ and STOI). Relatively, the logarithm compressed magnitude spectral loss does not benefit DEMUCS. Therefore, we reveal that DEMUCS can be further improved by properly revising the STFT terms of its loss function.",
language = "Chinese",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="dai-etal-2022-exploiting">
<titleInfo>
<title>Exploiting the compressed spectral loss for the learning of the DEMUCS speech enhancement network</title>
</titleInfo>
<name type="personal">
<namePart type="given">Chi-En</namePart>
<namePart type="family">Dai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qi-Wei</namePart>
<namePart type="family">Hong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jeih-Weih</namePart>
<namePart type="family">Hung</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2022-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<language>
<languageTerm type="text">Chinese</languageTerm>
<languageTerm type="code" authority="iso639-2b">chi</languageTerm>
</language>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 34th Conference on Computational Linguistics and Speech Processing (ROCLING 2022)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yung-Chun</namePart>
<namePart type="family">Chang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yi-Chin</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>The Association for Computational Linguistics and Chinese Language Processing (ACLCLP)</publisher>
<place>
<placeTerm type="text">Taipei, Taiwan</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>This study aims to improve a highly effective speech enhancement technique, DEMUCS, by revising the respective loss function in learning. DEMUCS, developed by Facebook Team, is built on the Wave-UNet and consists of convolutional layer encoding and decoding blocks with an LSTM layer in between. Although DEMUCS processes the input speech utterance purely in the time (wave) domain, the applied loss function consists of wave-domain L1 distance and multi-scale shorttime-Fourier-transform (STFT) loss. That is, both time- and frequency-domain features are taken into consideration in the learning of DEMUCS. In this study, we present revising the STFT loss in DEMUCS by employing the compressed magnitude spectrogram. The compression is done by either the power-law operation with a positive exponent less than one, or the logarithmic operation. We evaluate the presented novel framework on the VoiceBank-DEMAND database and task. The preliminary experimental results suggest that DEMUCS containing the power-law compressed magnitude spectral loss outperforms the original DEMUCS by providing the test utterances with higher objective quality and intelligibility scores (PESQ and STOI). Relatively, the logarithm compressed magnitude spectral loss does not benefit DEMUCS. Therefore, we reveal that DEMUCS can be further improved by properly revising the STFT terms of its loss function.</abstract>
<identifier type="citekey">dai-etal-2022-exploiting</identifier>
<location>
<url>https://aclanthology.org/2022.rocling-1.13</url>
</location>
<part>
<date>2022-11</date>
<extent unit="page">
<start>100</start>
<end>106</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Exploiting the compressed spectral loss for the learning of the DEMUCS speech enhancement network
%A Dai, Chi-En
%A Hong, Qi-Wei
%A Hung, Jeih-Weih
%Y Chang, Yung-Chun
%Y Huang, Yi-Chin
%S Proceedings of the 34th Conference on Computational Linguistics and Speech Processing (ROCLING 2022)
%D 2022
%8 November
%I The Association for Computational Linguistics and Chinese Language Processing (ACLCLP)
%C Taipei, Taiwan
%G Chinese
%F dai-etal-2022-exploiting
%X This study aims to improve a highly effective speech enhancement technique, DEMUCS, by revising the respective loss function in learning. DEMUCS, developed by Facebook Team, is built on the Wave-UNet and consists of convolutional layer encoding and decoding blocks with an LSTM layer in between. Although DEMUCS processes the input speech utterance purely in the time (wave) domain, the applied loss function consists of wave-domain L1 distance and multi-scale shorttime-Fourier-transform (STFT) loss. That is, both time- and frequency-domain features are taken into consideration in the learning of DEMUCS. In this study, we present revising the STFT loss in DEMUCS by employing the compressed magnitude spectrogram. The compression is done by either the power-law operation with a positive exponent less than one, or the logarithmic operation. We evaluate the presented novel framework on the VoiceBank-DEMAND database and task. The preliminary experimental results suggest that DEMUCS containing the power-law compressed magnitude spectral loss outperforms the original DEMUCS by providing the test utterances with higher objective quality and intelligibility scores (PESQ and STOI). Relatively, the logarithm compressed magnitude spectral loss does not benefit DEMUCS. Therefore, we reveal that DEMUCS can be further improved by properly revising the STFT terms of its loss function.
%U https://aclanthology.org/2022.rocling-1.13
%P 100-106
Markdown (Informal)
[Exploiting the compressed spectral loss for the learning of the DEMUCS speech enhancement network](https://aclanthology.org/2022.rocling-1.13) (Dai et al., ROCLING 2022)
ACL