In this paper, we explore a machine learning approach to evaluate audio quality for high sound pressure level (SPL) smartphone recordings. Our study is based on perceptual evaluations conducted by technical experts on eight audio sub-attributes (tonal balance, treble, midrange, bass, dynamics, temporal artifacts, spectral artifacts, and other artifacts) of audio quality for 121 smartphones released from 2019 to 2021. To address this task, we propose a Convolutional Neural Network (CNN) model, which proves to be a simple yet effective choice. We employ a pre-augmentation technique to enhance the training dataset size, creating a comprehensive dataset comprising recording spectrograms and corresponding perceptual evaluation scores. Our findings indicate that while the CNN model has certain limitations, it demonstrates promising capabilities in predicting evaluation scores, particularly in aspects of tonal balance, bass, and spectral artifact assessment.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.