The evaluation of audio quality is important in the development of immersive audio algorithms and reproduction systems, and binaural models are often used for this as a quick alternative to listening tests. Coloration (i.e., perceived loudness differences integrated across ears and frequency) is one key quality aspect; however, the majority of models used to predict coloration are often oversimplified or are missing a dedicated binaural stage to consider the relative contribution of the left and right ear signals. A binaural coloration model is presented that builds upon previous work and tests three different approaches for its binaural stage. The proposed model is evaluated in comparison with nine models that are frequently used to predict coloration by using data from five listening tests totaling 252 stimuli with various audio contents and source positions. The proposed model performed best with 85% of explained variance, followed by predictions based on ISO 532-1 loudness, yielding 78% explained variance. The commonly used log-spectral distance performed worst, with only 44% explained variance. The three tested binaural stages had little influence on the performance of the proposed model. The model is made freely available to download.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=22807
(593KB)
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 Join the AES. If you need to check your member status, login to the Member Portal.
