Within-Consonant Perceptual Differences in the Hearing Imired Ear Andrea Trevino* and Jont Allen HSR Research Group University of Illinois Urbana-Chamign
Objective: Undersnd how hearing imired (HI) ears perceive the acoustic cues in speech. Methods: 4 Consonant Vowel stimuli ( zero-error, unambiguous) /b, d, f, g, k, m, n, p, s,, t, v, ʒ, z/+/ɑ/ (in plots: Z= ʒ, S= ) 2 lkers for each consonant ( le, fele) Speech Weighted Noise at 4 SNRs (, 6, 2 db, Q) Token - lker & consonant Results: ) HI errors can be conined to a sll subset--- --- --- --------- of tokens ----------2) Tokens of the same consonant can be perceptually-- ------ ---different for HI listeners (noise-robustness/confusions) ----------3) Consistencies are observed across NH and HI ears in ------ ---terms of noise robustness and confusion groups
Audiogram and Average Error - 7 HI ears Threshold db HL Pure Tone Audiograms 2 4 6 8 2.25.25.5.5 2 3 4 6 8 Frequency khz g Avg P log P error error Average Probability of Error over all Stimuli 44L 44R 46L.5 46R 4L.25 NH 4R 36L 36R. 32L.5 32R 3L 3R 34L 34R. L R 22 6 2 6 4R SNR 2 Q 44L 44R 46L 46R 4L 4R 36L 36R 32L 32R 3L 3R 34L 34R L R 4R 4 Miller-Nicely Consonant Vowel stimuli, 2 lkers per consonant ( le, fele) /b, d, f, g, k, m, n, p, s,, t, v, ʒ, z/+/ɑ/, Speech Weighted Noise (SWN)
Audiogram and Average Error - 7 HI ears Threshold db HL 2 4 6 8 Pure Tone Audiograms 4L 2.25.25.5.5 2 3 4 6 8 Frequency khz g Avg Avg P log error P log error P error error Average Probability of Error over all Stimuli Average Probability of Error over all Stimuli 44L 44R 46L.5.5 46R 4L.25 NH 4R 36L 36R. 4L 32L 4L.5 32R 3L 3R 34L 34R.. L R 22 4R 22 6 6 2 2 6 2 2 Q SNR SNR L 44L R 44R 4R 46L 3L 46R 3R 4L 32L 4R 32R 36L 34L 36R 34R 32L 36L 32R 36R 3L 4L 3R 4R 34L 44L 34R 44R L 46L R 46R 4R Avg 4 Miller-Nicely Consonant Vowel stimuli 2 lkers per consonant ( le, fele) /b, d, f, g, k, m, n, p, s,, t, v, ʒ, z/+/ɑ/, Speech Weighted Noise (SWN)
Error Overview Probability of Error Probability of Error.8.6.4.2.8.6.4.2 Overview of Token Errors, Quiet, Listener 4L 6 db Quiet b b d d f g g k k mm n n p p s s t t v v S S Z Z z z Overview of Tokens, Errors, No Sorting Quiet, Listener 4L 6 db Quiet db d gd kg mg pk sk mm s t n t vn Sp Sp Zs zs kt p t Z v vz Sn Sn Zg m Z zb zv b f Sorted Tokens
Error Overview Probability of Error Probability of Error.8.6.4.2.8.6.4.2 Overview of Token Errors, Quiet, Listener 4L 6 db Quiet Probability of Error.8.6 b b d d f g g k k mm n n p p s s t t v v S S Z Z z z Overview of Tokens, Sorted Errors, No Tokens Sorting Quiet, Listener 4L.4 6 db Quiet.2 db d gd kg mg pk sk mm s t n t vn Sp Sp Zs zs kt p t Z v vz Sn Sn Zg m Z zb zv b f Sorted Tokens Overview of Token Errors, Listener 4L db 6 db 2 db Quiet Sorted Tokens
Probability of Error.8.6.4.2 Error Overview - 6 HI ears Overview of Token Errors, Listener 4L db 6 db 2 db Quiet Probability of Error.8.6.4.2 Overview of Token Errors, Listener 44L db 6 db 2 db Quiet Probability of Error.8.6.4.2 Overview of Token Errors, Listener 32R db 6 db 2 db Quiet Implications for undersnding an average. Or planning a training program. Sorted Tokens Sorted Tokens Sorted Tokens Probability of Error.8.6.4 Overview of Token Errors, Listener 36R db 6 db 2 db Quiet Probability of Error.8.6.4 Overview of Token Errors, Listener 46L db 6 db 2 db Quiet Probability of Error.8.6.4 Overview of Token Errors, Listener 34L db 6 db 2 db Quiet.2.2.2 Sorted Tokens Sorted Tokens Concentration of error to a subset of tokens observed across HI ears, in both Quiet and Noise Sorted Tokens
HI Within-Consonant Differences: Noise Robustness
.75.5.25 ba Probability of Error, Listener: 4L da ga fa Avg P error.75.5.25.75.5.25 va 62 Q na sa Sa Za za 62 Q 6 2 Q 62 Q SNR 62 Q N=4-8 per consonant 2 tokens per consonant
.75.5.25 ba Probability of Error, Listener: 4L da ga fa P error.75.5.25.75.5.25 va 62 Q na sa Sa 62 Q 62 Q SNR 62 Q Individual token differences in noise robustness Za za 62 Q Fele Male Avg Trevino, Allen, JASA 23 (In Review)
HI Within-Consonant Differences: Noise Robustness Average Error Difference, HI 4L.5 P e P e.5 g m t k S z p s n d v b Z Consonant Trevino, Allen, JASA 23 (In Review)
HI Within-Consonant Differences: Noise Robustness Average Error Difference, 7 HI Ears Mean Average Error Difference, HI 4L.5.5 P e P e.5.5 g m t k S z p s n d v b Z g m t k S z p s n d v b Z Consonant Consonant Differences in noise-robustness are observable across 7 HI ears Consistencies across HI ears suggest that each token's acoustic cues play a role Trevino, Allen, JASA 23 (In Review)
Characterizing Token Variability with Norl-Hearing (NH) psychoacoustic da Norl-hearing da SNR 9 Difference between tokens, NH Listeners NH SNR 9 [db] 5 5 g m t k S z p s n d v b Z Consonant The psychoacoustic noise-sked threshold (NH SNR9) is correlated with the intensity of the token s acoustic cue region (Régnier, 28). Comre the 2 tokens of each consonant. Trevino, Allen, JASA 23 (In Review)
HI vs. NH Noise Robustness Average Error Difference, 7 HI Ears Mean SNR 9 Difference between tokens, NH Listeners P e.5.5 NH SNR 9 [db] 5 5 g m t k S z p s n d v b Z Consonant g m t k S z p s n d v b Z Consonant Hearing Imired da Norl Hearing da Trevino, Allen, JASA 23 (In Review)
HI vs. NH Noise Robustness Implies that natural variability of cue intensity, characterized by Average Error Difference, 7 HI Ears SNR Difference between tokens, NH Listeners the 9 NH-listener noise-sking da, is strongly related to the Mean HI within-consonant differences in noise-robustness. P e.5.5 p-value<. NH SNR 9 [db] 5 5 g m t k S z p s n d v b Z Consonant g m t k S z p s n d v b Z Consonant Trevino, Allen, JASA 23 (In Review)
Conclusion There is a correlation between HI and NH consonant perception. Supports the hypothesis that the HI and NH listeners are using the same priry cues.
HI Within-Consonant Differences: Confusion groups
Token-Specific Confusions /bɑ/ Probability [%] Avg Recognition Da, all 7 HI ears, Fele /ba/ Token Avg Recognition Da, all 7 HI ears, Male /ba/ Token 8 b b 8 b d b d 6 f 6 f 5 g 5 g 4 k 4 k 3 m m n 3 v n 2 d p p s 2 s t t v v v sh sh g zh f zh z z p 5 6 2 Q SNR Probability [%] 5 6 2 Q SNR
Sumry Multiple tokens of the same consonant can be perceptually different for HI listeners in terms of noise-robustness and confusion groups Systetic behavior on a token basis: - HI and NH listeners show agreement in noise-robustness - Different HI listeners share token-dependent confusions Systetic responses suggest that common cues are being used Consistent (low-entropy) token-specific confusions imply that the HI ears are not guessing
Conclusions - The average error can have large amounts of underlying individual token error - Analysis at the token level reveals systetic tterns in the HI confusions - NH psychoacoustic da can characterize the cue variability of individual tokens, providing tools for undersnding HI perception - The role of naturally-occurring conflicting acoustic cues reins unknown (Li et al. 2, 22)
Thank you
All listeners, Prob of confusion for CV: ba Token-Specific Confusions /bɑ/ ll listeners, Prob of confusion for CV: ba f confusion for CV: ba Talker Talker All listeners, Prob of confusion for CV: ba All ba All listeners, da ba 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q da 6 2Q 6 2Q fa 6 2Q 6 R.8 Uttera 4R 3L 3R 32L 32R.8 34L 34R 36L 36Rfa 4L 4R ga 44L 44 ga.6.6 na 2L6 2Q32R 6 2Q34L 6 2Q34R 6 2Q36L 6 2Q36R 6 2Q4L 6 2Q4R 6 2Q 44L 6 2Q44R 46L 46R.4.4 na 34R 36L 36R 4L 4R 44L 44R 46L 46R sa.2.2 Talker sa 2 va Talker 2 6 2Q 6 2Q 6 2Q 6 2Q va 6 2Q 6 2QSa 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 Za 2Q 6 Sa 3R 32L Listener: L R 32R 34LListener: 34R L 4R 3L 3R 32L 36L R 36R 4R 4L 3L 4R 3R 44L 32L 44R za 32R Za 34 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 za SNR and Listener Error Probability Probability F Utter 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q M Utter M Utter.8.8 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q SNR and Listener d Listener 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q.6.6 Trevino, Allen, JASA 23 (In Review) Utter
All listeners, Prob of confusion for CV: ba ll listeners, Prob of confusion for CV: ba f confusion for CV: ba Talker Talker All listeners, Prob of confusion for CV: ba All ba All listeners, Prob of confusion for CV: ba All listeners, da ba 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q da Utterance 6 2Q 6 2Q fa 6 2Q 6 R 4R 3L 3R 32L 32R.8.8 34L 34R 36L 36Rfa 4L 4R ga ba 44L 44 da ga fa.6 ga.6 na 2L6 2Q32R 6 2Q34L 6 2Q34R 6 2Q36L 6 2Q36R 6 2Q4L 6 2Q4R 6 2Q 44L 6 2Q44R 46L 46R na Listener: L R 4R 3L 3R 32L.4 32R.4 34L 34R 36L 36R 4L 4R 44L na 44R 46L 46R 34R 36L 36R 4L 4R 44L 44R 46L 46R sa.2.8.2 Utterance 2 Talker sa 2 va.6 va xsa Talker 2 xza.4 6 2Q 6 2Q 6 2Q 6 2Q va 6 2Q 6 2QSa za 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6.2 Sa Za Listener: L R 4R 3L 3R 32L 32R Probability F Utter Probability M Utter Token-Specific Confusions /bɑ/ Error Probability Probability F Utter 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q Uttera 3R 32L 6 2Q 6 2Q32R 6 6 2Q34L 6 6 2Q34R 6 6 2Q36L 6 6 2Q36R 6 6 2Q4L 6 6 2Q Za4R 6 2Q 6 2Q44L 6 44R za 4 Listener: L R 4R 3L 3R 32L 3 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q SNR and 6 2Q Listener 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 za M Utter M Utter.8.8.6.6 SNR and Listener 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q SNR and Listener d Listener 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q Trevino, Allen, JASA 23 (In Review) Utter
All listeners, Prob of confusion for CV: ba ll listeners, Prob of confusion for CV: ba f confusion for CV: ba Talker Talker All listeners, Prob of confusion for CV: ba All listeners, Prob ba of c All listeners, Prob of confusion for CV: ba All ba listeners, Prob of da confu 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q da Utterance 6 2Q 6 2Q fa 6 2Q 6 R 4R 3L.8 3R 32L 32R 34L 34R 36L 36Rfa 4L 4R ga ba.8 44L 44 da ga fa ga.6 6 2Q 6 2Q 6 2Q.6 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 62Q 6 2Q s, Prob of confusion for CV: ba na 2L6 2Q32R 6 2Q34L 6 2Q34R 6 2Q36L 6 2Q36R 6 2Q4L 6 2Q4R 6 2Q 44L 6 2Q44R 46L 46R na Listener: L R.4 4R.43L 3R 32L 32R 34L 34R 36L 36R 4L 4R 44L na 44R 46L 46R 34R 36L 36R 4L 4R 44L Talker 44R 46L 46R sa.8.2.2 ba Utterance 2 Talker sa 2 va.6 da va xsa Talker 2 fa xza.4 va Sa za b of 6confusion 2Q 6for 2Q CV: ba 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 62Q 6 2Q 6 2Q ga6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6.2 Sa Za onfusion for CV: Listener: ba L R 4R 3L 3R 32L 32R 34L 34R Probability F Utter Probability M Utter Token-Specific Confusions /bɑ/ Error Probability Probability F Utter 3R 32L 6 2Q 6 2Q32R 6 6 2Q34L 6 6 2Q34R Talker 6 6 2Q 36L 6 2Q 6 2Q36R 6 6 2Q4L 6 6 2Q Za4R 6 2Q 6 2Q44L 6 44R za 4 Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q SNR and Listener 34L 34R 36L 36R 4L 4R 44L 44R 46L 46R M Utter Utter.8.8 SNR and Listener d Listener.6.6 Uttera Listener: L R 4R 3L 3R 32L 32R 34L 3 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 na 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 za Talker SNR and Listener ba Large proportion of /da, baga, fa va/ confusions 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6Talker 2Q 26 2Q 6 2Q 6 2Q 6 2Q 6 2Q da ga va 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6fa 2Q 6 2Q Sa ga Za 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q da sa Trevino, Allen, JASA 23 (In Review) Utter
Token-Specific Confusions /bɑ/ onfusion for CV: ba va 7% Probability [%] Talker Avg Recognition Da, all 7 HI ears, Fele /ba/ Token 8 b b d 6 f 5 g 4 k m 3 n p s t v v sh g Talker 2 zh z 4R 36L 36R 4L 2 4R 44L 44Rd 46L 46R 5 6 2 Q SNR ba da fa ga na sa va Sa Za za Talker : Large proportion of /da, ga, va/ confusions overall 7% ba 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q Listener Trevino, Allen, JASA 23 (In Review)
ga 6 2Q 6 2Q Token-Specific 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q Confusions 6 2Q 6 2Q ll listeners, Prob of confusion for CV: ba na /bɑ/ 4R 36L 36R 4L 4R 44L 44R 46L 46R Talker sa 3R 32L 32R 34L 34R Talker 36L 2 36R 4L va 4R 44L 44R ba 4 All listeners, Prob of confusion for CV: ba All listeners, All Sa da 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6Za 2Q 6 2Q 6 2Q fa 6 2Q 6 za R 4R 3L 3R 32L 32R.8.8 34L 34R 36L 36R 4L 4R ga 44L Uttera 44 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q.6.6 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 na 2L 2Q 32R 6 2Q 34L 6 2Q 34R 6 2Q 36L 6 2Q 36R 6 2Q 4L 6 2Q 4R 6 2Q 44L 6 2Q 44R 46L 46R Listener.4.4 sa.2.2 Talker 2 va 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2QSa 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 Za 2Q 6 SNR Listener: and Listener L R 4R 3L 3R 32L za 32R Uttera 3R 32L 32R 34LListener: 34R L 36L R 36R 4R 4L 3L 4R 3R 44L 32L 44R 34 SNR and Listener 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q SNR and Listener All listeners, Prob of confusion for CV: ba Error Probability Probability F Utter 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 M Utter M Utter.8.8.6.6 Trevino, Allen, JASA 23 (In Review) Uttera
ga 6 2Q 6 2Q Token-Specific 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q Confusions 6 2Q 6 2Q ll listeners, Prob of confusion for CV: ba na /bɑ/ 4R 36L 36R 4L 4R 44L 44R 46L 46R Talker sa 3R 32L 32R 34L 34R Talker 36L 2 36R 4L va 4R 44L 44R ba 4 All listeners, Prob of confusion for CV: ba All All listeners, Prob of c Salisteners, Prob of confu da 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6Za 2Q 6 2Q 6 2Q fa 6 2Q 6 za R 4R 3L.8.8 3R 32L 32R 34L 34R 36L 36R 4L 4R ga 44L Uttera 44 6 2Q 6 2Q 6 2Q.6 6 2Q.6 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 na 2L 2Q 32R 6 2Q 34L 6 2Q 34R 6 2Q 36L 6 2Q 36R 6 2Q 4L 6 2Q 4R 6 2Q 44L 6 2Q 44R 46L 46R Listener.4.4 sa.2.2 Talker 2 va 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2QSa 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 Za2Q 6 Listener: SNR L and RListener 4R 3L 3R 32L 32R 34L za34r Error Probability Probability F Utter Talker 2: Large proportion of /fa, va/ confusions overall Uttera 3R 32LListener: 32R L 34L R 34R 4R 36L 3L 36R 3R 4L 4R 32L 32R 44L 34L 44R 34 SNR and Listener M Utter M Utter.8.8 SNR and Listener.6.6 All listeners, Prob of confusion for CV: ba 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q Trevino, Allen, JASA 23 (In Review) Uttera
36L 36R 4L 4R 44L 44R 46L 46R Avg Recognition Da, all 7 HI ears, Male /ba/ Token Za 8 b b za d 6 f 5 g 4 k 6 2Q 6 2Q 6 2Q 63 2Q 6 2Q 6 2Q 6 2Q 6 2Q m v n p 2 s t v sh f zh z p ener Token-Specific Confusions /bɑ/ Probability [%] Talker 2 5 6 2 Q SNR ga na sa va Talker 2: Large proportion of /fa, va/ confusions overall Sa Trevino, Allen, JASA 23 (In Review)
All listeners, Prob of confusion for CV: ba ga Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q Token-Specific Confusions /bɑ/ na ll listeners, 32R 34L Prob 34R of confusion 36L for 36RCV: 4L ba 4R 44L 44R 46L 46R CV: ba sa Talker 4R 3L 3R 32L Talker 32R 34L 34R Talker 36L 236R 4L 4R 4 isteners, Prob of confusion All listeners, for CV: Prob ba of confusion for All CV: listeners, All listeners, ba Prob vaprob baof conf of ba Sa da 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q da 6 2Q 6 2Q 6 2Q 6 2QZa 6 2Q 6 Utterance fa R za 4R 3L 3R.8.8 32L 32R 34L 34Rfa.836L 36R 4L 4R ga 44L 44 ga.6.6.6 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 6 2Q 6 2Q 6 2Q 6.4 2Q 6 2Q 6 2Q 6 2Q 2Q 6 2Q 6 2Q 6 2Q na 2L 32R SNR 34L and Listener 34R.4 36L 36R 4L 4R na44l.4 44R 46L 46R 36R 4L 4R 44L 44R 46L 46R.2 sa.2 sa.2 Talker 2 6 2Q Talker 6 2Q 2 va 6 2Q 6 2Q Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 62Q 2Q 6 62Q 2Q 6 62Q va 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 62Q 6 62Q 2Q 6 62Q 2Q 6 62Q 2Q 6 Sa 62Q 2Q 6 Listener: L R 4R 32R 3L 34LListener: 3R 34R 32L L 36L 32R R 36RSNR 4R 34L 4R 4L Listener: and 3L Sa 34R Listener 3R 32L 32R 34LZa 34R 3L 4R L 36L 3R 44L R 36R 32L 44R 4R 4L 32R 46L 3L 4R 34L 46R Za 34 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q za 6 2Q 6 za SNR and Listener Error Probability Probability F Utter M Utter M Utter.8.8 SNR and Listener.6.6 Probability F Utter Analysis must be done at the token level 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q 6 2Q M Utter.8.6 Utterance 2 Trevino, Allen, JASA 23 (In Review)