Speech recognition plays an important role in a variety of applications for mobile communication. User communication devices for contact necessitate a broad vocabulary recognition scheme, greater precision, and a real-time, low-power schema. The power consumption and memory bandwidth of miniaturized battery-controlled devices are important. People's handheld devices often demand more effort, due to the speech challenge. As a result, a valuable technology based on the Stochastic Binary Cat Swarm Optimization Algorithm (SBCSO) is proposed in this research study to transform the non-audible murmur to normal voice. From the input murmured speech signal, the characteristics such as spectral skewness, spectral centroid, pitch chroma, and Taylor-Amplitude Modulation Spectrum are extracted and trained in the Deep Convolutional Neural Network (DCNN) classifier. The proposed stochastic binary cat swarm optimization algorithm is used to train DCNN classifier for speech recognition. To boost the results in metric analysis, the stochastic gradient descent algorithm and a Binary Cat Swarm Optimization Algorithm (BCSOA) are combined. In order to boost the experimental results in metric analysis, the stochastic gradient descent algorithm and BCSOA are combined in this research paper. The proposed algorithm performance is validated in terms of true positive rate, false positive rate and classification accuracy, and it showed better improved in speech recognition.