For internal use only / Copyright Siemens AG 2006. All rights reserved.
Contents Technical features Wind noise reduction 3 Automatic microphone system 9 Directional microphone system 15 Feedback cancellation 20 Page 2
Wind noise reduction Characteristics of wind noise Turbulences at microphone ports Highly fluctuating, non-stationary standard noise reduction algorithms (modulation based, Wiener filter etc.) do not work! Two microphone system: wind noise uncorrelated Correlation can be used for detection Page 3
Wind noise reduction TwinMic system BTE, 16 mm Speech correlated Wind noise uncorrelated Page 4
Wind noise reduction DM - Omni Filterbank Reduction of low frequency gain Σ Correlation detector control control Switch from directional mic to omni Reduce low frequency gain by up to 20 db Page 5
Wind noise reduction Technical measurements Page 6
Wind noise reduction Technical measurements 90 without WNR 80 with WNR 70 60 [d B] 50 40 30 20 100 frequency [Hz] Page 7
Wind noise reduction Summary wind noise reduction Standard noise reduction schemes can not reduce wind noise correlation of TwinMic signals can be used to detect wind noise Annoyance of wind noise is decreased by switching to omni and reducing low frequency gain Page 8
Automatic microphone matching Why do we need automatic microphone matching? Manufacturing tolerance of TwinMic pairs: phase 1, amplitude 1dB Aging: sensitivity (=amplitude) decreases by up to 2-3 db If amplitude mismatch > 0,5 db, DI decreases by ~ 1 db Phase mismatch leads to a maximum decrease in DI by 0,5-1 db Static matching during manufacturing (= starting point for automatic matching) Changes of microphone sensitivity after manufacturing should be compensated Automatic microphone matching of amplitude and phase Page 9
Automatic microphone matching Amplitude matching front mic filter bank BP 1 x adaptive directional processing BP 2 rear mic BP 3 BP 4 filter bank BP 1 BP 2 BP 3 BP 4 adaptive gain calc. Adaptation time constant: 16 s Range: +/- 4 db Four band filterbank (same as multichannel adaptivity) 0-350Hz, 350-800Hz, 800-2800Hz, 2800-10000Hz Automatic matching ensures depth of notch Adaptivity selects optimum direction for notch Page 10
Automatic microphone matching How many channels are required? Amplitude frequency response of unmatched microphone signals: (Example) frequency dependent mismatch 1 band matching: LP 0-2kHz 4 band matching ineffective! effective! Page 11
Automatic microphone matching Phase matching Variations of phase response due to aging can be modeled by a serial connection of two highpass filters Acoustic highpass : effects of mechanical part of microphone Electric highpass : effects of mechanical part of microphone Cut-off frequencies may vary due to production tolerances, aging and contamination ideal: fc_ac = 50Hz ideal: fc_mi = 180Hz ideal: fc_el = 100Hz simulates microphone and quantization noise simulates microphone sensitivity Page 12
Automatic microphone matching Effect of amplitude mismatch on directivity Mic1: gain = 0 db comp gain = -0.4 db Mic2: gain = -0.5 db Mic3: gain = 0 db comp gain = -0.4 db perfect (ideal): AI-DI: 7.36 mismatched (pract): AI-DI: 2.94 compensated (comp): AI-DI: 7.78 Page 13
Automatic microphone matching Effect of phase mismatch Mic1: 102 Hz (el) 50 Hz (ac) Mic2: 120 Hz (el) 50 Hz (ac) Mic3: 85 Hz (el) 50 Hz (ac) perfect (ideal): AI-DI: 7.36 mismatched (pract): AI-DI: 6.89 compensated (comp): AI-DI: 7.36 Page 14
Directional microphone system Reprise: broadband adaptive TwinMic speech speech speech noise noise noise Continuous adaptation of the directivity pattern to the current noise field Effective in particular for a single dominant noise source Page 15
Directional microphone system Block diagram adaptive (multichannel) TwinMic Filterbank Eight X 1-k 60 90 120 estimation of direction of noise incidence weighting factor k + 30 150 0 180 330 210 Cardiod 300 270 240 X Page 16
Directional microphone system Multichannel adaptivity: principle Directivity patterns are independently optimized in four bands 0-350Hz, 350-800Hz, 800-2800Hz, 2800-10000Hz Effective for more than 1 noise source (different spectra) Page 17
Directional microphone system Test condition Sentence tests, 20 hearing impaired subjects 0 Speech 270 90 HF Noise HF Noise Traffic Noise 180 omni static Multichannel adaptive Improvement in SRT: 1,5 db Page 18
Directional microphone matching Summary Aging causes phase and amplitude mismatch Amplitude mismatch reduces AI-DI by more than 4 db Phase mismatch reduces AI-DI by 0,5-1 db Automatic microphone matching in phase and amplitude ensures optimum directivity Multichannel adaptive directional mic can attenuate more than 1 noise simultaneously Improvement in SRT: ~ 1,5 db Page 19
Feedback cancellation Principle acoustical feedback path + - amplification adaptive filter Continuous estimation (!) of the feedback path impulse response by an adaptive filter subtraction of feedback components in the microphone signal no reduction of gain in critical frequency regions (main difference to notch filter approach) enhancement of sound quality near the critical gain (Pre Feedback) Challenge is optimum adaptation speed and filter length Page 20
Feedback cancellation Feedback paths depend on: Hearing aid type: BTE or ITE Vent size Obstacles around the head: hands, hats, telephones receivers, etc. Frequency response / db Page 21 0-20 -40-60 0 2 4 6 8 10-20 -40-60 0 2 4 6 8 10 0-20 ITE BTE open 2 mm 0.8 mm hand free -40-60 0 2 4 6 Frequency / khz 8 10 Static Highly time varying Hearing aid type Vent size Obstacles
Feedback cancellation Interindividual feedback paths (OLG) 90 80 70 60 50 40 30 20 10 10 2 10 3 10 4 3-4 khz region is most sensitive for feedback Maximum feedback-free gain can differ up to 20 db between individuals! Page 22
Feedback cancellation Adaptive phase cancellation (FBC) system is capable of instantaneously adjusting to varying acoustic environments L / db 120 110 100 90 80 70 60 + 10 db OLG - 10 db 50 100Hz 1kHz 10kHz FBC Off FBC On FBC Off FBC On FBC Off FBC On No whistling stable Whistling & smooth response Smooth & stable response Instable & peaky response Good sound quality Degraded sound quality, «reverberant» Page 23
Feedback cancellation Technical measurements 100 90 output level at critical gain critical gain FBC off critical gain FBC on 80 output level [db] 70 60 50 40 30 10 2 10 3 10 4 f [Hz] Page 24
Feedback cancellation Perceptual measurements Analysis for all subjects (n=6): Distribution of FBC off (+) vs. FBC on (-) 2.5 2 1.5 1 0.5 6 normal hearing subjects Which setting has a better sound quality? Critical gain 6 db (FBC on vs. Off) rating 0-0.5-1 -1.5-2 -2.5 speech elise flute glass No degradation of sound quality by FBC Page 25
Feedback cancellation Perceptual measurements Very clearly 20 hearing impaired subjects Do you hear feedback? Telephone receiver connected to CD player FirstFit telephone acoustically ENTdepartment, University of Munich Not at all Almost no feedback! Page 26
Feedback cancellation Perceptual measurements 6 normal hearing subjects Do you hear feedback? Telephone receiver connected to CD player 3mm vent size gain: without FBC, whistling clear audible In most cases feedback is eliminated Position of receiver is important ( counseling) Page 27
Feedback Cancellation Summary Adaptive filter continously estimates and cancels feedback path Feedback paths are not stationary >10 db variation! Eliminates feedback and improves sound quality 10-15 additional gain Recommendation: hearing aid should be stable at office without FBC (FBC is required for chewing, telephoning etc.) Page 28
Wind noise reduction Principle: Correlation Technical measurements: 20 db gain reduction Directional microphone system Automatic microphone matching: Ensures directivity Multichannel adaptivity: Benefit ~ 1,5 db SRT Feedback cancellation Principle: Adaptive filter Technical measurements: 10-15 db more gain Perceptual measurements: Almost no feedback on the phone Page 29