An optimized convolutional neural network for speech enhancement

Speech enhancement is an important property in today’s world because most applications use voice recognition as an important feature for performing operations in it. Perfect recognition of commands is achieved only by recognizing the voice correctly. Hence, the speech signal must be enhanced and free from background noise for the recognition process. In the existing approach, a recurrent convolutional encoder/decoder is used for denoising the speech signal. It utilized the signal-to-noise ratio property for enhancing the speech signal. It removes the noise signal effectively by having a low character error rate. But it does not describe the range of SNR of the noise added to the signal. Hence, in this, optimized deep learning is proposed to enhance the speech signal. AI function deep learning mimics the human brain's ability to analyze data and create patterns for use in making decisions. An optimized convolutional neural network was proposed for enhancing the speech for a different type of signal-to-noise ratio value of noises. Here, the particle swarm optimization process performs tuning the hyper-parameters of the convolutional neural network. The tuning of value is to minimize the character error rate of the signal. The proposed method is realized using MATLAB R2020b software and evaluation takes place by calculating the character error rate, PESQ, and STOI of the signal. Then, the comparison of the proposed and existing method takes place using evaluation metrics with − 5 dB, 0 dB, + 5 dB and + 10 dB. © 2023, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.