A METHOD FOR OPTIMIZING THE BASIC WAVELET FUNCTION BY THE CRITERION OF MINIMIZING THE ERROR OF APPROXIMATION OF SPEECH SIGNALS
DOI:
https://doi.org/10.18372/2310-5461.67.20244Keywords:
speech signals, adaptive wavelet features for speech recognition, wavelet transform, wavelet function optimization, minimization of approximation error, support vector method, genetic algorithmAbstract
The paper presents a developed method for selecting adaptive wavelet features for speech signal recognition based on optimizing the base wavelet function by the criterion of minimizing the approximation error. To build the base wavelet function, it was proposed to use a genetic algorithm, in which the accuracy of speech signal classification by the support vector method is the objective function. The main advantage of the constructed adaptive wavelet function using the proposed method is the fact that it takes into account the dynamics of the speech signal throughout time, since a wavelet function is built at each scale that depends on the entire signal and contains information about all its changes, which directly affects the improvement of the resolution of wavelet features of speech recognition. The analysis of speech signals by the proposed method will consist of the following steps: 1) determination of the speech signal metaclass; 2) selection of the optimal pair “adaptive wavelet function - classifier”; 3) wavelet analysis of the speech signal using the constructed adaptive wavelet function; 4) final classification of adaptive wavelet features for speech signal recognition. It is shown that the use of the basic wavelet functions generated on the basis of the presented method for the selection of adaptive wavelet features for speech signal recognition allows to obtain a steady increase in classification accuracy. In particular, in comparison with the mel-frequency capstral coefficients, the accuracy improvement ranges from 4.1 to 8.3 % for different phonemes of the Ukrainian language. It has been shown that the convolved adaptive wavelet function has the property of preserving information about dynamic processes in the speech signal, and, as a result, the dimensionality of adaptive wavelet recognition features is reduced by 1.5 times without loss of classification accuracy. It is shown that the use of the proposed method allows achieving better speech phoneme classification characteristics than using the classical analysis based on mel-frequency cepstral coefficients by a factor of 1.4.
References
K. T. Shivaram, N. Mahesh Kumar, M. Anusha and B. S. Manohar, “The optimal numerical wavelet based integration of probability density function by chebyshev wavelet method,” 2019 International Conference on Intelligent Computing and Control Systems (ICCS), Madurai, India, 2019, pp. 175-177, doi: 10.1109/ICCS45141.2019.9065515.
K. Geetha and M. K. Hota, “Seismic Random Noise Attenuation Using Optimal Empirical Wavelet Transform With a New Wavelet Thresholding Technique,” in IEEE Sensors Journal, vol. 24, no. 1, pp. 596-606, 1 Jan.1, 2024, doi: 10.1109/JSEN.2023.3334819.
C. Candemir, “Selection of Wavelet Based Optimal Denoising Method in fMRI Signals,” 2020 Innovations in Intelligent Systems and Applications Conference (ASYU), Istanbul, Turkey, 2020, pp. 1-5, doi: 10.1109/ASYU50717.2020.9259816.
O. Lavrynenko, B. Chumachenko, M. Zaliskyi, S. Chumachenko and D. Bakhtiiarov, “Method of Remote Biometric Identification of a Person by Voice based on Wavelet Packet Transform,” CEUR Workshop Proceedings, 2024, vol. 3654, pp. 150-162.
G. Yang, Y. Song and J. Du, “Speech Signal Denoising Algorithm and Simulation Based on Wavelet Threshold,” 2022 4th International Conference on Natural Language Processing (ICNLP), Xi'an, China, 2022, pp. 304-309, doi: 10.1109/ICNLP55136.2022.00055.
O. Lavrynenko, D. Bakhtiiarov, V. Kurushkin, S. Zavhorodnii, V. Antonov and P. Stanko, “A method for extracting the semantic features of speech signal recognition based on empirical wavelet transform,” Radioelectronic and Computer Systems, 2023, vol. 107, no. 3, pp. 101-124. doi: 10.32620/reks.2023.3.09.
Y. Tverdokhlib and V. Dubrovin, “Research on Wavelet Filter Features for Nonstationary Signals,” 2019 IEEE International Scientific-Practical Conference Problems of Infocommunications, Science and Technology (PIC S&T), Kyiv, Ukraine, 2019, pp. 785-788, doi: 10.1109/PICST47496.2019.9061501.
O. Veselska, O. Lavrynenko, R. Odarchenko, M. Zaliskyi, D. Bakhtiiarov, M. Karpinski and S. Rajba, “A Wavelet-Based Steganographic Method for Text Hiding in an Audio Signal,” Sensors, 2022, vol. 22, no. 15, pp. 1-25. doi: 10.3390/s22155832.
H. Xiao, D. Hu and J. Wang, “Threshold selection of wavelet denoising based on optimization algorithms,” 2022 International Conference on Innovations and Development of Information Technologies and Robotics (IDITR), Chengdu, China, 2022, pp. 88-92, doi: 10.1109/IDITR54676.2022.9796485.
S. Sun, Y. Sun, Y. Li, S. Ma, L. Zhang and Y. Hu, “An Adaptive Wavelet Multilevel Soft Threshold Algorithm for Denoising Partial Discharge Signals,” 2019 IEEE Sustainable Power and Energy Conference (iSPEC), Beijing, China, 2019, pp. 1874-1878, doi: 10.1109/iSPEC48194.2019.8975273.
O. Lavrynenko, R. Odarchenko, G. Konakhovych, A. Taranenko, D. Bakhtiiarov and T. Dyka, “Method of Semantic Coding of Speech Signals based on Empirical Wavelet Transform,” 2021 IEEE 4th International Conference on Advanced Information and Communication Technologies (AICT), Lviv, Ukraine, 2021, pp. 18-22, doi: 10.1109/AICT52120.2021.9628985.
O.Yu. Lavrynenko, D.I. Bakhtiiarov, B.S. Chumachenko, O.G. Holubnychyi, G.F. Konakhovych and V.V. Antonov, “Application of Daubechies wavelet analysis in problems of acoustic detection of UAVs,” CEUR Workshop Proceedings, 2024, vol. 3662, pp. 125-143.
J. Jiang, Z. Sun, R. Lu, L. Pan and Z. Peng, “Real Relative Encoding Genetic Algorithm for Workflow Scheduling in Heterogeneous Distributed Computing Systems,” in IEEE Transactions on Parallel and Distributed Systems, vol. 36, no. 1, pp. 1-14, Jan. 2025, doi: 10.1109/TPDS.2024.3492210.
Y. Yu et al., “Memristor Parallel Computing for a Matrix-Friendly Genetic Algorithm,” in IEEE Transactions on Evolutionary Computation, vol. 26, no. 5, pp. 901-910, Oct. 2022, doi: 10.1109/TEVC.2022.3144419.
M. Askari, S. Soleimani, M. H. Shakoor and M. Momeni, “Eliminating Network Depth: Genetic Algorithm for Parameter Optimization in CNNs,” in IEEE Access, vol. 13, pp. 22473-22488, 2025, doi: 10.1109/ACCESS.2025.3533976.
