Objective Measurement of Telephone Band Speech Quality Using Measuring Normalizing Blocks (MNBs)
|Publication Date:||1 August 1998|
Subjective quality assessment of speech codecs can be made in listening-only (one-way) tests or in conversational (two-way) tests. The objective quality measurement described in this American National Standard (ANS) estimates the subjective quality in listening-only tests of telephone band speech.1)
To demonstrate the subjective performance of a codec, the effects of a variety of quality factors should be investigated (see ITU-T Recommendation P.830). The accuracy of the objective quality measurement described in this ANS has not been verified for all of the factors specified in Recommendation P.830. Subclause 1.3 defines those conditions from P.830 to which this ANS applies.
When comparing a codec with another codec or with a reference condition based on subjective experimental results, statistical tests that take the distributions of subjective votes into account are often used. Since the objective measurement in this ANS estimates only the mean of subjective votes (e.g., MOS, DMOS), such statistical tests cannot be applied to the results of objective measurement. Prediction of percent poor or worse (%PoW) and percent good or better (%GoB) are currently under study.
The desirability of using objective measures to estimate user opinions of speech quality has long been a goal of the American National Standards Institute (ANSI) and the International Telecommunications Union, Telecommunication Standardisation Sector (ITU-T). Toward that end, the ITU-T adopted Recommendation P.861 in 1996. This Recommendation provides an algorithm that can be used under a limited set of conditions. These conditions are limited to waveform and CELP speech coding technologies under clear-channel conditions. With the proliferation of mobile telephone systems, it has become increasingly important to be able to predict user opinion under conditions of transmission channel errors and lower-rate speech coding. This American National Standard (ANS) defines an algorithm that provides acceptably accurate predictions in the same areas as Recommendation P.861, as well as in additional important conditions, such as transmission channel errors and lower-rate speech coders.
1) For purposes of this standard the passband is assumed to be 300-3400 Hz (-2 dB at 300 Hz; -3 dB at 3400 Hz), the same as shown in ITU-T Recommendation G.712 for PCM-derived channels between two 2-wire analog interfaces.