Non-Intrusive Estimation Model for the Speech-Quality Dimension Loudness

Inhalt:
In this article, we present an approach towards a new nonintrusive speech quality estimator. The proposed method facilitates the evaluation of speech telephony services and provides diagnostic information by assessing dimensions of the perceptual quality space. One of these quality dimensions is Loudness, which describes a non optimal sound level. As an important part of the proposed model, a non-intrusive Loudness estimator is presented. The estimator uses a linear regression with five different indicators that are extracted from the output signal only, to map subjective Loudness judgments. The new model is trained on one and tested on two independent subjective databases. In addition, the performance of the Loudness estimator is compared to the diagnostic intrusive quality estimator Diagnostic Intrusive Assessment of Listening quality (DIAL). The evaluation shows that the estimator provides results on a high reliability level, indicating the applicability and the value of the proposed estimator for diagnostic enhancement.