Table of Contents
Advertisement
C H A P T E R
2 :
T E S T
M E T H O D O L O G Y ?
based measurement algorithms assume that any audible difference between the
two input signals is a distortion, and by definition the "enhanced" signal will
sound different than the unprocessed signal.
To get around this, a clean signal as the reference file (R) is recommended. This
shall be distorted artificially, which results in signal D, the distorted reference
signal. Signal D may now be used as the input to the enhancer. The output of
the enhancer will be E, the enhanced signal.
When assessing speech quality at this time, the clean reference R and the
enhanced signal E as the input signals of OPERA should be chosen. The grade
calculated by the measurement algorithm now indicates how similar the
enhanced signal sounds to the clean reference. This also implies that based on
the measurement no statement can be made weather the enhanced signal
sounds better than the original signal or not.. If the measurement result shall be
compared to the result obtained from a listening test, it is important to
remember that the question to the subjects must be "How much does the
enhanced signal differ from the reference signal?" and not "Does the enhanced
signal sound better or worse than the reference signal?". There is no
standardized measure available today which answers the second question,
which is frequently a matter of personal taste.
Going one step further, calculate the gain achieved by the enhancer when
relating the final MOS derived this way, to the MOS achieved by comparing the
clean reference (R) to the distorted reference (D). Figure 2.5 shows a sketch of
such a setup.
Figure 2.5: Setup for measuring signal enhancers.
21
Advertisement
Table of Contents
