Assessing the effectiveness of transfer learning strategies in BLSTM networks for speech denoising

Denoising speech signals represent a challenging task due to the increasing number of applications and technologies currently implemented in communication and portable devices. In those applications, challenging environmental conditions such as background noise, reverberation, and other sound artif...

Full description

Saved in:
Bibliographic Details
Main Authors: Marvin Coto-Jiménez, Astryd González-Salazar, Michelle Gutiérrez-Muñoz
Format: Article
Language:English
Published: Instituto Tecnológico de Costa Rica 2022-11-01
Series:Tecnología en Marcha
Subjects:
Online Access:https://172.20.14.50/index.php/tec_marcha/article/view/6448
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850161726678368256
author Marvin Coto-Jiménez
Astryd González-Salazar
Michelle Gutiérrez-Muñoz
author_facet Marvin Coto-Jiménez
Astryd González-Salazar
Michelle Gutiérrez-Muñoz
author_sort Marvin Coto-Jiménez
collection DOAJ
description Denoising speech signals represent a challenging task due to the increasing number of applications and technologies currently implemented in communication and portable devices. In those applications, challenging environmental conditions such as background noise, reverberation, and other sound artifacts can affect the quality of the signals. As a result, it also impacts the systems for speech recognition, speaker identification, and sound source localization, among many others. For denoising the speech signals degraded with the many kinds and possibly different levels of noise, several algorithms have been proposed during the past decades, with recent proposals based on deep learning presented as state-of-the-art, in particular those based on Long Short-Term Memory Networks (LSTM and Bidirectional-LSMT). In this work, a comparative study on different transfer learning strategies for reducing training time and increase the effectiveness of this kind of network is presented. The reduction in training time is one of the most critical challenges due to the high computational cost of training LSTM and BLSTM. Those strategies arose from the different options to initialize the networks, using clean or noisy information of several types. Results show the convenience of transferring information from a single case of denoising network to the rest, with a significant reduction in training time and denoising capabilities of the BLSTM networks.
format Article
id doaj-art-f99a5b276b2d41a2b1bf04ffe7d473f1
institution OA Journals
issn 0379-3982
2215-3241
language English
publishDate 2022-11-01
publisher Instituto Tecnológico de Costa Rica
record_format Article
series Tecnología en Marcha
spelling doaj-art-f99a5b276b2d41a2b1bf04ffe7d473f12025-08-20T02:22:44ZengInstituto Tecnológico de Costa RicaTecnología en Marcha0379-39822215-32412022-11-0135810.18845/tm.v35i8.6448Assessing the effectiveness of transfer learning strategies in BLSTM networks for speech denoisingMarvin Coto-JiménezAstryd González-SalazarMichelle Gutiérrez-Muñoz Denoising speech signals represent a challenging task due to the increasing number of applications and technologies currently implemented in communication and portable devices. In those applications, challenging environmental conditions such as background noise, reverberation, and other sound artifacts can affect the quality of the signals. As a result, it also impacts the systems for speech recognition, speaker identification, and sound source localization, among many others. For denoising the speech signals degraded with the many kinds and possibly different levels of noise, several algorithms have been proposed during the past decades, with recent proposals based on deep learning presented as state-of-the-art, in particular those based on Long Short-Term Memory Networks (LSTM and Bidirectional-LSMT). In this work, a comparative study on different transfer learning strategies for reducing training time and increase the effectiveness of this kind of network is presented. The reduction in training time is one of the most critical challenges due to the high computational cost of training LSTM and BLSTM. Those strategies arose from the different options to initialize the networks, using clean or noisy information of several types. Results show the convenience of transferring information from a single case of denoising network to the rest, with a significant reduction in training time and denoising capabilities of the BLSTM networks. https://172.20.14.50/index.php/tec_marcha/article/view/6448BLSTMdeep learningtransfer learningspeech processing
spellingShingle Marvin Coto-Jiménez
Astryd González-Salazar
Michelle Gutiérrez-Muñoz
Assessing the effectiveness of transfer learning strategies in BLSTM networks for speech denoising
Tecnología en Marcha
BLSTM
deep learning
transfer learning
speech processing
title Assessing the effectiveness of transfer learning strategies in BLSTM networks for speech denoising
title_full Assessing the effectiveness of transfer learning strategies in BLSTM networks for speech denoising
title_fullStr Assessing the effectiveness of transfer learning strategies in BLSTM networks for speech denoising
title_full_unstemmed Assessing the effectiveness of transfer learning strategies in BLSTM networks for speech denoising
title_short Assessing the effectiveness of transfer learning strategies in BLSTM networks for speech denoising
title_sort assessing the effectiveness of transfer learning strategies in blstm networks for speech denoising
topic BLSTM
deep learning
transfer learning
speech processing
url https://172.20.14.50/index.php/tec_marcha/article/view/6448
work_keys_str_mv AT marvincotojimenez assessingtheeffectivenessoftransferlearningstrategiesinblstmnetworksforspeechdenoising
AT astrydgonzalezsalazar assessingtheeffectivenessoftransferlearningstrategiesinblstmnetworksforspeechdenoising
AT michellegutierrezmunoz assessingtheeffectivenessoftransferlearningstrategiesinblstmnetworksforspeechdenoising