|
Hi
All, I
have a question about the confidence interval (CI) of the objective audio
quality. Of course, CI should be a concept with the subjective quality scores.
However, once we compute the objective score using one audio quality metric
(such as PEAQ, PESQ, NMR, whatever), how much is this objective score believable
with respect to the real quality, supposing we don?t know the real quality. For
example, if the objective PEAQ score is -1.5, can we believe the magnitude of
the real quality is [-2~-1]? I
tried to compute the CI by using the standard samples for PEAQ as following: the
objective quality (ODG) is computed for each sample firstly, and then the
difference (D) between ODG and subjective quality (SDG) is gotten, for all
samples. I assume the difference D is approximately normally distributed, so the
CI (e.g. 95%) of D can be computed according to the T-distribution table,
standard deviation of D, and sample amount. For example, for PEAQ, the computed
95% CI is 0.16. Then, if the ODG of one sequence is computed by using PEAQ, can
we think the subjective quality SDG should belong to the interval [ODG-0.16,
ODG+0.16] with the probability 95%? Is this way to compute the CI of objective
quality right? I
really appreciate any comment about this question, thank you so
much. Best
regards, Junyong You |