Computational acoustic testing has significant benefits for aberrant speech diagnosis, surveillance, and early diagnosis. This work provides a feature extracting mechanism for vocal pathology identification using singular value decomposition (SVD) datasets. Speech noise detection techniques rely on the precision of acoustic measurements, which are used to evaluate voice health. Then, a system made up of 27 neural network layers—consisting of recurrent and convolutional neural networks—processes the characteristics extracted (RNN). The dataset is divided into test and training sets using a tenfold validation method, allowing the system’s efficacy to be assessed. According to the experimental findings, the CNN model’s estimated accuracy is 87.11%, whereas the RNN model’s is 86.52%. The cross-validation of tenfold is another method used to evaluate the efficacy of the classifier. The Tensor Flow library is used in the Python development of the system’s program, and one NVidia Titan X GPU is used for computing. The operating system used for all calculations is Linux. Healthcare practitioners working in the field of vocal analysis and diagnostic will find the suggested technique to be a useful tool in the early detection and surveillance of voice disorders.