1. Objective intelligibility measurements (3)

[Lorenzo Picinali <LPicinali@xxxxxxxxx>]

Dear Matt,
I don't know if you have ever been using PESQ (Perceptual Evaluation of the Speech Quality, ITU Recommendation P.862)...
We started using it three years ago for the objective evaluation of hearing aids audio quality, but at the end we concluded that for that specific task, PESQ can generate problems...
Nevertheless, it seems that for your problems it could be worth trying: actually, PESQ should give an estimation of the MOS (Mean Opinion Score, ITU-T P.800) for the quality of the speech, and not specifically for the intelligibility, but there is an interesting paper where speech intelligibility measurements have been done using a modified version of PESQ (the authors were Beerends, Larsen, Iyer ad van Vugt "Measurement of speech intelligibility based on the PESQ approach"...if you are interested, I can send you the article)...
The PESQ MOS estimation is done comparing two signals, a reference one and the signal under test: you could estimate the MOS (or MOS-LQO, which is a sort of the remapping of the PESQ score in order to make it closer to the MOS) for two couples, correspondent to the two communication system to be tested with and without added noise, i.e.:
-Couple1: System1WithoutNoise - System1WithNoise
-Couple2: System2WithoutNoise - System2WithNoise
Getting the two scores, you should be able to have an estimation on which of the two system is more "resistant" to additive noise...
I'm not sure whether PESQ can be considered as perfectly suitable for your task, but I really think it could be worth trying!
If you need any more information about PESQ and about how does it work, I can just send you a brief analytical description!
Yours
Lorenzo

--
Lorenzo Picinali
PhD Student
Music, Technology and Innovation Research Centre
0116 2551551, internal 6770
Clephan Building CL0.19
De Montfort University
Leicester