Definition of the Speech Reception Threshold (SRT)

Aging and Speech Communication: Peripheral and Cognitive Factors A personal account on several decades of research on factors underlying individual differences in SRT in noise. Tammo Houtgast and Sophia Kramer VU University medical center Amsterdam, The Netherlands The European scene: Germany, The Netherlands: Strong tradition in hearing research, dominated by physicists and engineers. Great Britain, Scandinavia (i.e. Sweden): more involvement of psychologists. In The Netherlands, it took a long time before cognitive aspects in relation to speech comprehension were taken seriously. Bloomington, October 2007 Bloomington, October 2007 Definition of the Speech Reception Threshold (SRT) The typical engineering approach: measure the SRT in noise for a large and diverse group of subjects; use the SII to account for each individual s hearing threshold (audibility); the remaining variance in SII must be related to other factors than the pure-tone audiogram. Searching for auditory test that explain (part of) the remaining variance. What tests? How much variance explained? sentences correct (%) 100 80 60 40 20 0-14 -12-10 -8-6 -4 0 2 4 speech-to-noise ratio (db) Note: this relates to speech-shaped stationary noise. Bloomington, October 2007 D-term Normal-hearing From Plomp Note: Normal hearing in noise >30 db: SRT at -5dB speech-to-noise ratio Plomp 1986 Note: SRT in noise only weakly related to SRT in quiet 1

SRT in stationary noise and in fluctuating noise: SRT in steady noise (db S/N) SRT fluct 10 8 6 4 2 0 NH more sensitive -4-6 -8-10 -12-14 -16-16 -14-12 -10-8 -6-4 0 2 4 6 8 10 SRT in fluctuating noise (db S/N) HI The D-term in the SRT in noise also includes the effect of, for instance, a high-frequency hearing loss. We like to differentiate between: - simple hearing loss, as reflected by the PTA (audibility problem only) - complex hearing loss (also supra-threshold deficits) From SRT to the corresponding SII SRT Kramer 2001 Speech area: the relevant dynamic range 1/3 of this range is required for 50% intelligibility (SRT) 30 db dynamic range Mean long-term speech spectrum Noise spectrum with S/N is -5 db For normal hearing: SRT at 5 db S/N ratio With high-frequency hearing loss: With high-frequency hearing loss: a better S/N ratio is required to meet the 1/3 rule The SRT increases 2

Normal hearing Hearing impaired with high fr. loss Speech-area at SRT (50% sentences) SII SRT (Note: in reality SII-calculations are somewhat more complicated) Remember: SII SRT includes effect of the audiogram (audibility). Normal hearing: SRT SII SRT = 0.33 Simple hear. impaired: higher SRT, but also with SII SRT = 0.33 Complex hear. impaired: higher SRT, with SII SRT >> 0.33 : implies sub-optimal supra-threshold processing. SRT for sentences with low or high context: predictability of each successive letter Why the concentration on peripheral auditory tests? Two older studies relevant to the question [cognition and SRT]: Van Rooij and Plomp 1991: Is the observed decline of SRT with age related to a decline of cognitive performance? Smoorenburg 1992: A detailed statistical analysis of a large body of data on SRT s for subjects with noise-induced hearing loss. SRT for sentences with low or high context: predictability of each successive letter SRT for sentences with low or high context: predictability of each successive letter 3

SRT for sentences with low or high context: predictability of each successive letter SRT for sentences with low or high context: predictability of each successive letter SRT for sentences with low or high context: predictability of each successive letter SRT for sentences with low or high context: predictability of each successive letter SRT for sentences with low or high context: predictability of each successive letter (mean age 25.8) Noise level lower educational level (mean age 80.6) higher educational level (mean age 81.6) Mean entropy Low redundancy High predictability High redundancy Low predictability van Rooij and Plomp, JASA 1991 Hypothesis: the benefit from context requires extensive top-down processing. No interaction between the effects of age and linguistic entropy on the SRT. Their conclusion: the elderly show no decline in cognitive performance. Note: This concerns the SRT in stationary noise! 4

Smoorenburg JASA 1992 : 400 ears with various degrees of noise-induced hearing loss; SRT in quiet and in stationary noise; Tone audiogram. Basic question: relation between SRT and tone audiogram Smoorenburg, JASA 1992 Smoorenburg, JASA 1992 Smoorenburg, JASA 1992 Smoorenburg JASA 1992: 400 ears with various degrees of noise-induced hearing loss: SRT in quiet and in stationary noise Tone audiogram is important predictor of SRT N. No other relevant peripheral factors found. Thus, cognition! In search of other predictors of SRT N, besides the PTA: auditory bandwidth excessive upward spread of masking Peripheral factors recruitment dead regions (avant le lettre *) AI- model calculations (similar to the SII model) cognitive skills *) A limited array of completely defective receptors Smoorenburg, JASA 1992 5

His final conclusion: SRT N is determined by the tone audiogram and cognitive skill! An interesting approach to estimate the amount of a nonperipheral factor involved in SRT N differences among a group of subjects: The correlation between left-ear and right-ear SRT s. SII SRT (i.e., audibility accounted for) Left ear Smoorenburg, JASA 1992 No suggestions on how this cognitive skill in speech perception can be measured. Right ear Note: This concerns a special group: middle aged (under 55), with noise-induced HL. This approach was not really adopted in hearing research with the elderly. SII SRT (i.e., audibility accounted for) No left-right correlation: no evidence of variance in SRT-related cognitive factors. The continued search for peripheral factors related to SRT in noise 10 NH listeners Further correlation studies (S and T test), young and older subjects: spectral acuity temporal acuity 1000 10 800 8 The distortion-sensitivity approach (the individual approach) ERB [Hz] 400 measurement error ERD [ms] 4 The latest effort: Temporal jitter 200 100 20 30 40 50 60 70 80 90 PDR [%] 2 measurement error 1 20 30 40 50 60 70 80 90 PDR [%] PDR [%]: Position in Dynamic Range Hilkhuyzen 2004 Spectral resolution (ERB in Hz) 450 400 350 300 250 200 150 100 50 NHR SIM-S SIM-F HI-S HI-F Level-dependency of S and T Spectral resolution 55 60 65 70 75 80 85 90 Presentation level (db SPL) Temporal resolution (ERD in ms) 16 14 12 10 8 6 4 2 0 NHR SIM-S SIM-F HI-S HI-F Temporal resolution 55 60 65 70 75 80 85 90 Presentation level (db SPL) Note: only T qualifies as a supra-threshold deficit George 2005 SRT-fluc vs. temporele resolutie (T) SRT-modulated [db SNR] -4-6 -8-10 -12-14 -16-18 2 4 6 8 10 12 14 16 18 Temporal resolution [ERD in ms] : normal hearing George 2005 6

The continued search for peripheral factors related to SRT in noise Distortion-Sensitivity approach: poor Further correlation studies (audiogram, S and T test), young and older subjects: SRT-predictors: - The pure-tone audiogram - Temporal resolution -Age( ) performance impaired normal The distortion-sensitivity approach (the individual approach) The latest effort: Temporal jitter Impaired: (without glasses) Normal: (with glasses) good original distortion Picture out of focus (distorted) Distortion-Sensitivity approach: Example of the distortion-sensitivity approach poor performance impaired normal good original distortion SRTa in noise [db S/N ratio] 12 10 8 6 4 2 0 12 12 a b c 10 8 8 6 6 4 4 2 2 0 0 0 10 20 0 1 2 3 4 5 6 7 0,0 0,2 0,4 0,6 0,8 1,0 intensity perturbation [db] temporal perturbation [ # wavelets] spectral perturbation [oct.] 10 Hypothesis: the hearing impaired is less sensitive to that type of external distortion which corresponds to the internal distortion causing the poor performance. Also: The detection threshold for that type of external distortion will be poorer (because it has to compete with the same type of internal distortion). Van Schijndel, 1998 Note: very accurate SRT-measurements are required for conclusive data. Also: The application of external distortion in one domain exclusively is not possible; there is always some degree of collateral damage. The continued search for peripheral factors related to SRT in noise Further correlation studies (audiogram, S and T test), young and older subjects: SRT-predictors: - The pure-tone audiogram - Temporal resolution -Age( ) Binaural processing requires accurate temporal coding of the fine structure BILD: SRT-difference between N 0 S 0 and N 0 S pi The distortion-sensitivity approach (the individual approach): - Very accurate data required. - Always some degree of collateral distortion in other domains. The latest effort: Temporal jitter (n=6) (n=17) (n=8) Goverts 2004 7

Distortion-sensitivity approach: The effect of external distortion on the BILD The BILD as a function of external jitter of the fine structure NH NH Goverts 2004 Next: detection threshold and effect on SRT in noise. For precise manipulation of the fine-structure jitter we used Introducing various degrees of phase-jitter: Wavelet-analysis: Log(f) Log(f) t (s) A (intensity, amplitude 2 ) PHASE (finestructure, carrier) t (s) A (intensity, amplitude 2 ) PHASE (finestructure, carrier) Wavelet-analysis Phase randomization (jitter) Re-synthesis scale 0... 1.0 1.0 is noisy speech ( Shannon-speech ) Jitter detection threshold in speech (NH only) Jitter detection threshold in speech (low-pass) filter condition detection threshold (% jitter) 1 NH-y, n = 10 NH-e, n = 10 HI-e, n = 25 Control (broadband low-pass high-pass 7.7% 8.9% 15.7% *# fine-structure dist. thresh. (jitter) 0.1 Table 1. Effects of filtering on detection threshold for NH-y subjects (n = 10). Low-pass filter <1.3 khz, high-pass filter >1.3 khz. Significant difference from control group is indicated with *, significant difference from low-pass condition is indicated with #. 0.01 Test/retest st. error is factor 1.19 For 15/25 HI-subjects det.threshold > 2 times standard deviation for NH 8

SRT s for NH subjects as a function of the degree of jitter SRT stat SRT fluct The continued search for peripheral factors related to SRT in noise Further correlation studies (audiogram, S and T test), young and older subjects: SRT-predictors: - The pure-tone audiogram - Temporal resolution -Age( ) The distortion-sensitivity approach (the individual approach): - Very accurate data required. - Always some degree of collateral distortion in other domains. The latest effort: Temporal jitter - Seems interesting, needs more work (perhaps this meting). Future: Apply the distortion-sensitivity approach to the SRT with HI-subjects. Combine within one group of subjects with BILD data. Recapitulation at this point The European project HearCom Co-variance between left- and right-ear SRT: provides estimate of how much non-peripheral (cognitive) aspects are involved. Only T qualifies as a supra-threshold deficit, not S. Distortion-sensitivity approach interesting but labour-intensive. Typically 60% of variance in SRTN is explained by peripheral tests. Aspects of temporal fine-structure jitter need to be worked out. EU 6th Framework: Information Society Technologies (IST), e-inclusion. An Integrated Project: Hearing in the communication society HEARCOM Returning to my introduction: What have the beta s in Europe (physicists, engineers) learned? Duration sept 2004 - march 2009, budget 9 mln Euro. About 25 partners from Universities, Research Inst., Industry. Coordination: VU University medical center Amsterdam EFAS 2007 HearCom main objectives To achieve full participation in the information society For hearing impaired people For all in adverse conditions (acoustics, interfering noises, telecommunications,.) For the improvement of Participation in social life Labor participation Quality of life Safety HEARCOM Reference-diagram Personal SP1 Hearing Problem Deficiencies Solution Communication Environment Adverse Ambient Acoustics Rehabilitation Assistive Applications eservices (Internet) EFAS 2007 9

HearCom SP1: Hearing Deficiencies Screening tests on communication performance 3-digits SRT test by telephone (Dutch, German, English, Swedish, French?, ), later to be implemented on Internet. OLSA for different languages? Discussions on a test on spatial hearing / localization Diagnostics Impairment tests / the auditory profile Concensus on the tests to be included for clinical use (also cognitive tests!) SRT-modulated [db SNR] -4-6 -8-10 -12-14 -16 SRT-fluc vs. TRT-bars non-auditory! auditory -18 52 54 56 58 60 62 64 66 TRT-bars [% text] : normal hearing George 2005 SRT-fluc vs. both temporal resolution and TRT-bars (explains 70-80% of the variance in SRT-mod) Cognitive functioning as an additional explanatory factor in intelligibility in noise George 2005 Hearing and speech perception in the presence of background noise; underlying components involved: Peripheral auditory Central auditory disentangle modality-specific auditory processes from more global cognitive functioning Cognitive / Linguistic Psychological (Humes,1995,1996, 2007; Pichora-Fuller et al.,1995; Zhao and Stephens, 2000;Gatehouse et al.,2003; Lunner et al., 2003; Hallgren,2005) (McFarland and Cacace, 1995; Watson et al., 1996; Pichora-Fuller et al., 2005; Humes et al., 2007) 10

Speech-Reception-Threshold in noise * List of short everyday Dutch sentences Two types of masks (stationary, 16 Hz modulated-block) Adaptive procedure SRT: threshold at which 50% correctly reproduced Prozacis Indiana s finest product * (Plomp & Mimpen, 1979 b) Prozac is Indiana s finest product Prozac isindiana s finest product Prozac is Indiana sfinest product Prozac is Indiana sfinest product 11

Visual analogue 28% unmasked text 34% 40% 46% 52% 58% 64% 70% TRT score: percentage unmasked text needed to read 50% of the sentences correctly Two experiments 1. Normally hearing Correlation SRT en TRT 2. Hearing impaired (and normally hearing) Regression analysis with Speech-in-Noise as the dependent variable Experiment 1 The relationship between the Speech-Reception-Treshold in noise (SRT) and the Text-Reception-Threshold (TRT: a visual analogue) Participants, normally hearing (N=34) Gender: 24 females, 10 males Ages: M=34 yrs (sd 18.4), 19 78 years PTA at 0.25, 0,5,1,2kHz: PTA at 4 khz: < 15 db HL < 30 db HL 12

Procedure Two Blocks 3 x TRT 1 x SRTstat 2 x SRTmod Result (normally hearing subjects) R = 0.54 (p < 0.01) R = 0.54 (p < 0.01) Experiment 2, hearing impaired and normally hearing Test battery Hearing loss [db HL] 0 20 40 60 80 Gehoorverlies 100.125.25.50 1 2 4 8 Frequency [khz] Normally hearing Hearing impaired N 13 21 Gender male / female 5/8 9/12 Age range, mean 53 78, 63.5 46 81, 65.5 Pure-tone audiometry 3 x Text-Reception-Threshold test (TRT) 2 x Speech-Reception-Threshold, modulated noise (SRTmod) Temporal Acuity Spectral Acuity Regression Analysis Result Experiment 2 (Normally hearing, N=13) Dependent (to predict) variable: Speech-Reception-in-Noise test (SRTmod) Correlations SRTmod p Independent variables (predictors): Pure-tone audiometry Temporal acuity Spectral acuity Text-Reception-Test (TRT) Age Pure-tone Audiometry Temporal acuity Spectral acuity Text-Reception-Threshold (TRT) test Age 0.48-0.09 0.22 0.60 0.43 p < 0.01 13

Result Experiment 2 (Hearing impaired, N=21) Stepwise multiple regression analysis Non-verbal cognitive tests Cambridge Neuropsychological Testing Automated Battery (CANTAB) Language independent (Predict the outcome on the Speech-in-Noise test) Predicotor Cum R² Multipele R Rapid Visual Processing Step 1 Temporal Acuity 0.48 0.85 Spatial Working Memory Step 2 Text-Reception-Test 0.73 Rapid Visual Processing (RVP), sustained attention Spatial Working Memory RVP-d (sensitivity to target sequence) and RVP-ß (tendency to respond regardless of presence of target sequence) SWM-errors (number of times a used box is opened) and SWM-strategy (efficient search strategy, i.e. following a predetermined search sequence) SRT as function of TRT Correlations Cantab, SRT, TRT (N=60) -4 R =. 60 SRT TRT R =.43, p < 0.01 n.s n.s R =.37, p < 0.01 working memory sustained attention SRT-modulated [db SNR] -6-8 -10-12 -14 non-auditory auditory -16 NHR : normally hearing HI : hearing impaired -18 52 54 56 58 60 62 64 66 TRT-bars [% text] 14

Conclusions More details: Zekveld AA, Deijen JB, Goverts ST, Kramer SE (2007). The relationship between nonverbal cognitive About 30% of the interindividual variance in both the TRT and SRT appears to be associated with modality aspecific cognitive and linguistic skills TRT may be clinically relevant to determine part of the origin (auditory or non-auditory) of deteriorated speech reception among hearing impaired listeners functions and hearing loss. J Speech Lang. Hear. Research, 50: 74-82. Zekveld AA, George ELJ, Kramer SE, Goverts ST, Houtgast T (2007). The development of the text reception threshold test: a visual analogue of the speech reception threshold. J Speech Lang. Hear. Research, 50: 576-84. George ELJ, Zekveld AA, Kramer SE, Goverts ST, Festen JM, Houtgast T (2007). Auditory and nonauditory factors affecting speech reception in noise by older listeners. J. Acoust. Soc. Am., 121(4): 2362-75. Thanks for your attention! 15