Application of neural networks for prediction of subjectively assessed interior aircraft noise

Products are increasingly judged by their acoustic performance; and during the last decades, sound quality in general has gained a lot of attention, both from academia and companies. An obstacle in the evaluation of the sound quality is that jury testing is time consuming and require human resources. In an attempt to overcome these limitations, neural networks have been applied in this work with the objective to find a relation between human acoustic perception and a quantity possible to physically measure. For this purpose, 30 of 170 sound samples of interior aircraft noise have been subjectively assessed during jury testing with 40 participators. With extracted psychoacoustic features from the sound samples and the obtained results from the jury testing, a shallow neural network (SNN) with one hidden layer is trained and tested. The prediction performance of the SNN is compared with another alternative method - multiple linear regression. The evaluation of the remaining un-assessed sound samples is predicted by the trained SNN and implemented in the deep learning neural networks, such as Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN)