Efficient Gated Convolutional Recurrent Neural Networks for Real-Time Speech Enhancement
Deep learning (DL) networks have grown into powerful alternatives for speech enhancement and have achieved excellent results by improving speech quality, intelligibility, and background noise suppression. Due to high computational load, most of the DL models for speech enhancement are difficult to i...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Universidad Internacional de La Rioja (UNIR)
2025-01-01
|
| Series: | International Journal of Interactive Multimedia and Artificial Intelligence |
| Subjects: | |
| Online Access: | https://www.ijimai.org/journal/bibcite/reference/3324 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1841560762102841344 |
|---|---|
| author | Fazal-E -Wahab Zhongfu Ye Nasir Saleem Hamza Ali Imad Ali |
| author_facet | Fazal-E -Wahab Zhongfu Ye Nasir Saleem Hamza Ali Imad Ali |
| author_sort | Fazal-E -Wahab |
| collection | DOAJ |
| description | Deep learning (DL) networks have grown into powerful alternatives for speech enhancement and have achieved excellent results by improving speech quality, intelligibility, and background noise suppression. Due to high computational load, most of the DL models for speech enhancement are difficult to implement for realtime processing. It is challenging to formulate resource efficient and compact networks. In order to address this problem, we propose a resource efficient convolutional recurrent network to learn the complex ratio mask for real-time speech enhancement. Convolutional encoder-decoder and gated recurrent units (GRUs) are integrated into the Convolutional recurrent network architecture, thereby formulating a causal system appropriate for real-time speech processing. Parallel GRU grouping and efficient skipped connection techniques are engaged to achieve a compact network. In the proposed network, the causal encoder-decoder is composed of five convolutional (Conv2D) and deconvolutional (Deconv2D) layers. Leaky linear rectified unit (ReLU) is applied to all layers apart from the output layer where softplus activation to confine the network output to positive is utilized. Furthermore, batch normalization is adopted after every convolution (or deconvolution)
and prior to activation. In the proposed network, different noise types and speakers can be used in training and testing. With the LibriSpeech dataset, the experiments show that the proposed real-time approach leads to improved objective perceptual quality and intelligibility with much fewer trainable parameters than existing LSTM and GRU models. The proposed model obtained an average of 83.53% STOI scores and 2.52 PESQ scores, respectively. The quality and intelligibility are improved by 31.61% and 17.18% respectively over noisy speech. |
| format | Article |
| id | doaj-art-2d97090989654187b23f20ecf8dd03b3 |
| institution | Kabale University |
| issn | 1989-1660 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | Universidad Internacional de La Rioja (UNIR) |
| record_format | Article |
| series | International Journal of Interactive Multimedia and Artificial Intelligence |
| spelling | doaj-art-2d97090989654187b23f20ecf8dd03b32025-01-03T15:20:35ZengUniversidad Internacional de La Rioja (UNIR)International Journal of Interactive Multimedia and Artificial Intelligence1989-16602025-01-0191667410.9781/ijimai.2023.05.007ijimai.2023.05.007Efficient Gated Convolutional Recurrent Neural Networks for Real-Time Speech EnhancementFazal-E -WahabZhongfu YeNasir SaleemHamza AliImad AliDeep learning (DL) networks have grown into powerful alternatives for speech enhancement and have achieved excellent results by improving speech quality, intelligibility, and background noise suppression. Due to high computational load, most of the DL models for speech enhancement are difficult to implement for realtime processing. It is challenging to formulate resource efficient and compact networks. In order to address this problem, we propose a resource efficient convolutional recurrent network to learn the complex ratio mask for real-time speech enhancement. Convolutional encoder-decoder and gated recurrent units (GRUs) are integrated into the Convolutional recurrent network architecture, thereby formulating a causal system appropriate for real-time speech processing. Parallel GRU grouping and efficient skipped connection techniques are engaged to achieve a compact network. In the proposed network, the causal encoder-decoder is composed of five convolutional (Conv2D) and deconvolutional (Deconv2D) layers. Leaky linear rectified unit (ReLU) is applied to all layers apart from the output layer where softplus activation to confine the network output to positive is utilized. Furthermore, batch normalization is adopted after every convolution (or deconvolution) and prior to activation. In the proposed network, different noise types and speakers can be used in training and testing. With the LibriSpeech dataset, the experiments show that the proposed real-time approach leads to improved objective perceptual quality and intelligibility with much fewer trainable parameters than existing LSTM and GRU models. The proposed model obtained an average of 83.53% STOI scores and 2.52 PESQ scores, respectively. The quality and intelligibility are improved by 31.61% and 17.18% respectively over noisy speech.https://www.ijimai.org/journal/bibcite/reference/3324convolutional gated recurrent unit (convolutional gru)deep learningintelligibilitylong short term memory (lstm)speech enhancement |
| spellingShingle | Fazal-E -Wahab Zhongfu Ye Nasir Saleem Hamza Ali Imad Ali Efficient Gated Convolutional Recurrent Neural Networks for Real-Time Speech Enhancement International Journal of Interactive Multimedia and Artificial Intelligence convolutional gated recurrent unit (convolutional gru) deep learning intelligibility long short term memory (lstm) speech enhancement |
| title | Efficient Gated Convolutional Recurrent Neural Networks for Real-Time Speech Enhancement |
| title_full | Efficient Gated Convolutional Recurrent Neural Networks for Real-Time Speech Enhancement |
| title_fullStr | Efficient Gated Convolutional Recurrent Neural Networks for Real-Time Speech Enhancement |
| title_full_unstemmed | Efficient Gated Convolutional Recurrent Neural Networks for Real-Time Speech Enhancement |
| title_short | Efficient Gated Convolutional Recurrent Neural Networks for Real-Time Speech Enhancement |
| title_sort | efficient gated convolutional recurrent neural networks for real time speech enhancement |
| topic | convolutional gated recurrent unit (convolutional gru) deep learning intelligibility long short term memory (lstm) speech enhancement |
| url | https://www.ijimai.org/journal/bibcite/reference/3324 |
| work_keys_str_mv | AT fazalewahab efficientgatedconvolutionalrecurrentneuralnetworksforrealtimespeechenhancement AT zhongfuye efficientgatedconvolutionalrecurrentneuralnetworksforrealtimespeechenhancement AT nasirsaleem efficientgatedconvolutionalrecurrentneuralnetworksforrealtimespeechenhancement AT hamzaali efficientgatedconvolutionalrecurrentneuralnetworksforrealtimespeechenhancement AT imadali efficientgatedconvolutionalrecurrentneuralnetworksforrealtimespeechenhancement |