Signal speech reconstruction and noise removal using convolutional denoising audioencoders with neural deep learning

Abouzid, Houda; Chakkor, Otman; Reyes, Óscar; Ventura, Sebastián

doi:10.1007/s10470-019-01446-6

Cited by 22 publications

(9 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, artificial neural networks, especially autoencoders, have attracted attention in functional genomics for their ability to fill-in missing data for image restoration and inpainting ( Chaitanya et al, 2017 ; Ghosh et al, 2020 ; Mao et al, 2016 ; Xie et al, 2012 ). Autoencoders are neural networks tasked with the problem of simply reconstructing the original input data, with constraints applied to the network architecture or transformations applied to the input data in order to achieve a desired goal like dimensionality reduction or compression, and de-noising or de-masking ( Abouzid et al, 2019 ; Liu et al, 2020 ; Voulodimos et al, 2018 ). Stochastic noise or masking is used to modify or remove data inputs, training the autoencoder to reconstruct the original uncorrupted data from corrupted inputs ( Tian et al, 2020 ).…”

Section: Introductionmentioning

confidence: 99%

Rapid, Reference-Free human genotype imputation with denoising autoencoders

Dias

Evans

Chen

et al. 2022

eLife

View full text Add to dashboard Cite

Genotype imputation is a foundational tool for population genetics. Standard statistical imputation approaches rely on the co-location of large whole-genome sequencing-based reference panels, powerful computing environments, and potentially sensitive genetic study data. This results in computational resource and privacy-risk barriers to access to cutting-edge imputation techniques. Moreover, the accuracy of current statistical approaches is known to degrade in regions of low and complex linkage disequilibrium. Artificial neural network-based imputation approaches may overcome these limitations by encoding complex genotype relationships in easily portable inference models. Here we demonstrate an autoencoder-based approach for genotype imputation, using a large, commonly used reference panel, and spanning the entirety of human chromosome 22. Our autoencoder-based genotype imputation strategy achieved superior imputation accuracy across the allele-frequency spectrum and across genomes of diverse ancestry, while delivering at least 4-fold faster inference run time relative to standard imputation tools.

show abstract

Section: Introductionmentioning

confidence: 99%

Rapid, Reference-Free human genotype imputation with denoising autoencoders

Dias

Evans

Chen

et al. 2022

eLife

View full text Add to dashboard Cite

show abstract

“…To solve this problem and enhance the signal quality recorded in real-life environments, several algorithms have been developed throughout the years. In this work, a method based on deep neural networks (DNN) was proposed to map the noisy speech to clean speech [4,5]. As mentioned, with degraded signals, speech technologies do not work properly, for this reason, the DNN approach can be implemented for better results in several applications, such as in mobile phone applications, speech recognition systems, and assistive technology [6,7].…”

Section: Introductionmentioning

confidence: 99%

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

Coto-Jiménez¹,

González-Salazar²,

Gutiérrez-Muñoz³

2022

View full text Add to dashboard Cite

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.

show abstract

“…Unlike other deep AE methods, CNN uses convolutional filters rather than neurons to extract the required feature map. Recently, CAE is utilized in many applications, e.g., radar-based activity classification [27], denoising of speech signals [28], and fault detection in aircraft engine [29]. In the geophysical community, CAE solves enormous problems, such as, lithology prediction [30], arrival picking [31], seismic data interpolation [32], simultaneous-source separation [33], earthquake parameters classification [34], and waveform-based sourcelocation imaging [35].…”

Section: Introductionmentioning

confidence: 99%

Seismic Data Compression Using Deep Learning

et al. 2021

View full text Add to dashboard Cite

The exponential growth of the size of seismic data recorded in seismic surveys and real time data monitoring makes seismic data compression strongly demanded. Furthermore, compression will lead to an efficient use of the bandwidth assigned for the communication link between the seismic stations and the main center. In this paper, two convolutional autoencoders (CAEs) are proposed for seismic data compression. The two algorithms are mainly based on the convolutional neural network (CNN), which has the capability to compress the seismic data into feature representations, thereby allowing the decoder to perfectly reconstruct the input seismic data. The results show that the first model is efficient at low compression ratios (CRs), while the second model improves the signal-to-noise ratio (SNR) from about 3 dB to 12 dB compared to the other benchmark algorithms at moderate and high CRs. INDEX TERMS Convolutional autoencoders (CAE), Deep learning, Seismic data compression.

show abstract

Signal speech reconstruction and noise removal using convolutional denoising audioencoders with neural deep learning

Cited by 22 publications

References 15 publications

Rapid, Reference-Free human genotype imputation with denoising autoencoders

Rapid, Reference-Free human genotype imputation with denoising autoencoders

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

Seismic Data Compression Using Deep Learning

Contact Info

Product

Resources

About