Comparative performance of an adaptive directional microphone and a multi-channel noise reduction system, in a digital hearing instrument Kevin CP Yuen Institute of Human Communicative Research & Division of Otolaryngology, Department of Surgery, The Chinese University of Hong Kong Anna CS Kam Phonak Hong Kong Hearing Centre Polly SH Lau Division of Speech and Hearing Sciences, The University of Hong Kong 8 Oct 2005

2 Objectives  Compare the speech recognition in noise performance among Multi-channel noise reduction system (medium) (FNC), Multi-channel noise reduction system (medium) (FNC), Adaptive directional microphone (DAZ), & Adaptive directional microphone (DAZ), & FNC + DAZ FNC + DAZ  Compare the objective and subjective performance between NAL-NL1 in Perseo, & NAL-NL1 in Perseo, & Digital Perception Processing (DPP) in Perseo Digital Perception Processing (DPP) in Perseo

3 Multi-channel noise reduction system - Fine noise canceller, FNC (Medium) Courtesy of Phonak  Frequency- dependent gain reduction  SNR-dependent gain reduction

4 Adaptive directional microphone (DAZ)  Polar patterns are switched adaptively in response to the direction of noise Courtesy of Phonak

5 NAL-NL1 & DPP  NAL-NL1, National Acoustics Laboratory – Non-linear 1 (Dillion, 1999) Provides gain-frequency response that maximizes speech intelligibility for any input level of speech, Provides gain-frequency response that maximizes speech intelligibility for any input level of speech, while keeping overall loudness at a level no greater than that perceived by a normal hearing person, or at a lower level if it provides greater speech intelligibility. while keeping overall loudness at a level no greater than that perceived by a normal hearing person, or at a lower level if it provides greater speech intelligibility.  DPP, Digital Perception Processing Incorporates a psychoacoustic loudness mode of normal and impaired cochlea function (Moore et al, 2003) Incorporates a psychoacoustic loudness mode of normal and impaired cochlea function (Moore et al, 2003) Calculates excitation patterns and specific loudness patterns evoked in a normal and impaired cochlear, from the incoming signal, and adjusts gain required to match the loudness patterns of impaired cochlea to those of normal cochlea Calculates excitation patterns and specific loudness patterns evoked in a normal and impaired cochlear, from the incoming signal, and adjusts gain required to match the loudness patterns of impaired cochlea to those of normal cochlea Aims at restoring loudness perception to ‘normal’ Aims at restoring loudness perception to ‘normal’

6 Research Design Group 1 (5 subjects) OWN hearing aid → Perseo DPP→Perseo NAL-NL1 1 week 1 week 1 week 1 week Group 2 (4 subjects) OWN hearing aid → Perseo NAL-NL1 → Perseo DPP Perseo DPP = Digital Perception Processing target Perseo NAL-NL1 = NAL-NL1 target

7 Assessment tools  DPP vs NAL-NL1 / FNC+DAZ vs DAZ vs FNC Cantonese Hearing-in-Noise Test (CHINT) Cantonese Hearing-in-Noise Test (CHINT)  (Wong & Soli, 2005) Chinese translated version of Abbreviated Profile of Hearing Aid Benefit (APHAB) Chinese translated version of Abbreviated Profile of Hearing Aid Benefit (APHAB)  (Cox & Alexander, 1995)  DPP vs NAL-NL1 Sound quality paired comparison judgment Sound quality paired comparison judgment

8 Subjects  Nine post-lingually deafened adults ( yrs)  Sensorineural hearing loss  Native Cantonese-speaking  Hearing aid experience > 6 months Own hearing aid Perseo hearing aid Multiband noise reduction Directional mic Compression scheme Ear S1OCL211 S2OCR211 S3OCR211 S4 ✔ WDRCR311 S5OCL311 S6 ✔✔ WDRCR211 S7WDRCL211 S8WDRCL211 S9 ✔✔ WDRCR211 Average unaided air-conduction threshold of the test ear (+/- 1 SD)

9 Speech recognition in noise results  Separately for Signal_Front/ Noise_Front & Signal_Front/ Noise_Front & Signal_Front/ Noise_Side Signal_Front/ Noise_Side  Run Two-way repeated measures ANOVA for Noise suppression (FNC+DAZ vs DAZ vs FNC) Noise suppression (FNC+DAZ vs DAZ vs FNC) Prescription (DPP vs NAL-NL1) Prescription (DPP vs NAL-NL1)& Post-hoc Tukey HSD test

10 Results Noise suppression options FNC+DAZ vs DAZ vs FNC

11S1S2S3S4S5S6S7S8S9SNR FNC+DAZ vs DAZ FNC+DAZ vs FNC DAZ vs FNC FNC+DAZ vs DAZ vs FNC Signal_Front / Noise_Front

12S1S2S3S4S5S6S7S8S9SNR FNC+DAZ vs DAZ DAZ FNC+DAZ vs FNC FNC+ DAZ FNC+DAZ DAZ vs FNC DAZDAZDAZDAZDAZDAZDAZDAZ FNC+DAZ vs DAZ vs FNC Signal_Front / Noise_Side

13 Results Prescriptions DPP vs NAL-NL1

14S1S2S3S4S5S6S7S8S9SNR DPP vs NAL-NL1DPP DPP vs NAL-NL1 Signal_Front / Noise_Front

15S1S2S3S4S5S6S7S8S9SNR DPP vs NAL-NL1DPPDPPDPPDPP DPP vs NAL-NL1 Signal_Front / Noise_Side

16 APHAB DPP vs NAL-NL1 APHAB subscale Sums of negative ranks (NAL-NL1 > DPP) Sums of positive ranks (DPP > NAL-NL1) p value Ease of communication Reverberation Background noise Adversiveness Wilcoxon Signed-Ranks Test

17 Sound quality paired comparison judgment DPP vs NAL-NL1 SOFT loudness MEDIUM loudness so1 Bird singing m1 Alarm clock so2Catm2 Bird singing so3 Clock ticking m3 Church bell so4 Distant thunder m4 Classical music so5 Running water m5Raining so6 Male radio broadcast m6 Toilet flushing so7 Female radio broadcast m7 Male radio broadcast so8 Female singing m8 Female radio broadcast so9 Male singing m9 Female singing m10 Male singing

18 Sound quality paired comparison judgment DPP vs NAL-NL1 Track Subject So 1 So 2 So 3 So 4 So 5 So 6 So 7 So 8 So 9 M1M1 M2M2 M3M3 M4M4 M5M5 M6M6 M7M7 M8M8 M9M9 M 10 S1 -DNDNDNDNDDDDDDNDDN S2 DNDNDDDDDNDDDNDNNND S3 NNNDDDDDD-DDNNDDDNN S4 DDNDNNDDN-DDDDDNDDN S5 --DNNDDDD-NDDNDDDNN S6 DNDNNNDDN-DDDNNNDDN S7 NNDNDDNDN-D-NNNDNND S8 DNDDNDNNN-DDNNNNDDN S9 DNDNNNDNN-NNDNNDNND

19 Sound quality paired comparison judgment DPP vs NAL-NL1 ND S1NNNNNNDDDDDDDDDDDD S2NNNNNNNDDDDDDDDDDDD712 S3NNNNNNNDDDDDDDDDDD S4NNNNNNDDDDDDDDDDDD S5NNNNNNDDDDDDDDDD S6NNNNNNNNNDDDDDDDDD - 99 S7NNNNNNNNNNDDDDDDD S8NNNNNNNNNNDDDDDDDD S9NNNNNNNNNNNNDDDDDD ND DPP NAL Wilcoxon Signed-Ranks Test, p =.26

20 Real ear SPL difference DPP minus NAL-NL1 DPP Difference, dB

21 Conclusion  Speech recognition in noise With signal and noise from the front, With signal and noise from the front,  DPP better than NAL-NL1 in 1 subject  FNC, DAZ and FNC+DAZ were not different With signal from the front, noise from the side With signal from the front, noise from the side  DPP better than NAL-NL1 in 4 subjects  DAZ better than FNC in 8 subjects  FNC+DAZ better than FNC in 6 subjects Whenever there was a difference, DPP was better than NAL-NL1 Whenever there was a difference, DPP was better than NAL-NL1 Adaptive Directional Mic better than Multiband noise reduction system, only for signal_front/noise_side but not for signal_front/noise_front Adaptive Directional Mic better than Multiband noise reduction system, only for signal_front/noise_side but not for signal_front/noise_front  best speech recognition, measured AI-weighted directional pattern for signal_front/noise_side (Ricketts et al, 2002)

22 Conclusion  APHAB NAL-NL1 > DPP (Adversiveness) NAL-NL1 > DPP (Adversiveness) DPP > NAL-NL1 (Reverberation) DPP > NAL-NL1 (Reverberation)  Sound quality paired comparison judgment 5/9 subjects had more preferences towards DPP than NAL-NL1 5/9 subjects had more preferences towards DPP than NAL-NL1 3/9 subjects had more preference towards NAL-NL1 than DPP 3/9 subjects had more preference towards NAL-NL1 than DPP

23 NAL-NL1 English does not maximize speech intelligibility for Cantonese ? Band importance function Band importance function  English sentence – peak at 2000Hz (Eisenberg et al, 1998)  Cantonese sentence – peak at Hz (Chua, 2004) Band importance function -> derive SII -> derive NAL-NL1 Band importance function -> derive SII -> derive NAL-NL1 Chua, 2004

24 Kevin Yuen Division of Otorhinolaryngology, Department of Surgery The Chinese University of Hong Kong

25 2cc coupler gain response of the three hearing aid settings (front 0 degree azimuth)