Vehicle interior sound quality evaluation model based on convolutional neural network

The in-vehicle interior sound quality evaluation model, leveraging deep learning methods, minimizes reliance on acoustic theory and empirical knowledge. It efficiently extracts intricate features and objectively aligns evaluation results with subjective perceptions, achieving high efficacy. To align with human auditory perception and facilitate feature extraction in deep learning, the collected noise samples were preprocessed using the Mel frequency cepstrum coefficient. Subsequently, a convolutional neural network-based sound quality evaluation model was constructed. With adequate training and appropriate hyperparameters, the model attained a training accuracy of 92.16%. Noise samples were then utilized to validate the model, yielding a verification accuracy of 93.33%. Further validation through a confusion matrix demonstrated that the model's overall prediction evaluation grade closely aligns with the actual evaluation grade, confirming a strong correlation between the model's predictions and subjective evaluation results.