Proceedings Fonetik 2009 - Institutionen för lingvistik

Proceedings, FONETIK 2009, Dept. of Linguistics, Stockholm UniversityTable 2. Description of the hearing impaired testsubjects groupsSwedish German Flemish# Subject 15 15+15 15+15HearingImpairmentModerateMild+ModerateModerate+CochlearImplantsLocation KTH- Hörtech- Viataal-Preliminary AnalysisMean results of the SRT measurement testsare presented in Figure 2. The figure shows thelevel of SRT value for the different hearing impairmentgroups (cochlear implants with a noticeablyhigher level than the other groups), aswell the difference in SRT value with andwithout using SynFace. The mean values do notshow significant decrease or increase in theSRT level when using SynFace than with audio-onlyconditions. Nevertheless, when lookingat the performance of the subjects individually,a high inter-subject variability is clearwhich means that certain subjects have benefitedfrom the use of SynFace. Figure 3 showsthe sorted delta SRT value per subject for theSwedish moderate hearing impaired subject andthe Dutch cochlear implants subjects in speechwith babble noise condition. In addition to thehigh variability among subjects, and the highrange scaling between the groups with differenthearing impairment levels, it is clear that, in thecase of babble noise, most of the Swedish moderatehearing impairments subjects show benefit(negative delta SRT).Regarding the results of the effort scaling,subjects at all locations, do not show significantdifference in scaling value between the conditionof speech with and speech without Syn-Face. But again, the scaling value shows a highinter-subject variability.Another investigation we carried out was tostudy the effect of the SRT measurement listlength on the SRT value. As mentioned before,the SRT measurement used lists of 20 sentences,where every sentence contained 5words, and one training measurement was doneat the beginning to eliminate any training effect.Still, when looking at the average trend ofthe SRT value over time for each sentence, theSRT value was decreasing, this can be explainedas an ongoing training throughout themeasurement for each subject. But when lookingat the individual SRT value per test calculatedafter the 10 th and the 20st sentence foreach measurement, an observation was that forsome of the measurements, the SRT value ofthe same measurement increased at the 20 st sentencecompared to the 10 th sentence. Figure 4presents the difference of SRT value at the 20stsentence and the 10 th sentence for 40 SRT measurementswhich shows that although most ofthe measurements had a decreasing SRT value,some of them had an increasing one. Thismeans that the longer measurement is not alwaysbetter (decreasing the learning effect).We suspect here that this can be a result ofthat the 20 sentences long measurements aretoo long for the hearing impaired subjects, andthat they might be getting tired and loosingconcentration when the measurement is as longas 20 sentences and hence requiring a higherSNR.Synface, Sentence test252015SRTs105Icra without SynfaceIcra with SynfaceBabble without SynfaceBabble with Synface0-5Sweden;KTH, mod HIGemany;HTCH, mildHIGemany;HTCH, modHINetherlands,Viataal, modHINetherlands,Viataal, CIFigure 2. Mean SRT value for each of the subjects groups with and without the use of SynFaceand with two types of noise: stationary and babble.142