Towards Estimation of Quality of Watermarked Audio Signal Using Objective Measures

This paper compares two objective audio quality assessment measures calculated for three watermark methods with its corresponding subjective quality. The aim was to see if these measures could be used to estimate the subjective audio quality with various watermarks. Samples were watermarked with the LSB substitution, the direct spread-spectrum, and the echo hiding methods. The objective scores were calculated using peaqb, an implementation of the ITU-R BS.1387-1 standard, and PEMO-Q. PEMO-Q showed significantly higher correlation, about 0.90 compared to peaqb. Initial quality estimation tests were also conducted, where regression from objective score to the subjective score of one watermark (e.g. LSB) was estimated, and this regression was used to estimate the subjective score of another watermark method (e.g. spread-spectrum) from its objective score. PEMO-Q showed higher estimation accuracy, with Root Mean Square Error (RMSE) of about 11%.