On Mitigating Acoustic Feedback in Hearing Aids with Frequency Warping by All-Pass Networks Presented at the 178 th Acoustical Society of America Meeting, December 2019 Harinath Garudadri hgarudadri@ucsd.edu +1 858 668 6128 Qualcomm Institute of Calit2 University of California, San Diego
To enable psychophysical investigations beyond what is possible today – Miller and Donahue Open Speech Signal Processing Platform Workshop, NIH, Bethesda, MD, Oct. 2014 2 / 16
Pisha, et al., (2019), IEEE Access
OSP Development Model ADB UART H A Linux Browser RT-MHA Embedded Web Server Communication Enabled F User Command Line B C Devices Interface 802.11 802.15 Web Apps Reference WiFi PAN (BT) C React Laravel Designs xMAC 4G/5G D Node.js SQLite HTML (experimental) /PHP libOSP libDSP xPhy GPS NginX (experimental) Processing and Daemon for syncing Communication Drivers with AWS E E Device (PCD) ALSA IIO SPI Custom E FPGA F G E Wireless Transducer FPGA FPGA FPGA Multi-channel Devices (TDs) Sensing & BTE-RICs ePhys IMU Actuation Users and Clinical Web Developers and Signal Processing B C A D Researchers DSP Engineers Researchers Embedded Computing Wireless & Comm Computer F H E G and Wearables Researchers Scientists 4 / 16
OSP Development Model ADB UART H A Linux Browser RT-MHA Embedded Web Server Communication Enabled F User Command Line B C Devices Interface 802.11 802.15 Web Apps Reference WiFi PAN (BT) C React Laravel Designs xMAC 4G/5G D Node.js SQLite HTML (experimental) /PHP libOSP libDSP xPhy GPS NginX (experimental) Processing and Daemon for syncing Communication Drivers with AWS E E Device (PCD) ALSA IIO SPI Custom E FPGA F G E Wireless Transducer FPGA FPGA FPGA Multi-channel Devices (TDs) Sensing & BTE-RICs ePhys IMU Actuation Users and Clinical Web Developers and Signal Processing B C A D Researchers DSP Engineers Researchers Embedded Computing Wireless & Comm Computer F H E G and Wearables Researchers Scientists 5 / 16
Freping – A portmanteau for Fre quency War ping Discrete Representation of Signals, Allpass Network Oppenheim and Johnson, IEEE Proceedings, 1972. v ( − n ) (1 − α 2 ) z − 1 z − 1 − α z − 1 − α 1 1 − α z − 1 1 − α z − 1 1 − α z − 1 1 − α z − 1 q 0 ( n ) q 1 ( n ) q 2 ( n ) q 3 ( n ) f f f f Realtime frequency warping Freping Frequency- Framing Original All-pass Overlap-add warped & signal network ( α ) windowing signal 6/16
When do Hearing aids howl? Nyquist Stability Criteria (NSC) due to acoustic feedback � ⌘� ⇣ F ( e j ω , n ) − ˆ � � G ( e j ω , n ) F ( e j ω , n ) � � ≥ 1 , (magnitude cond.) � � ⇣ ⌘ F ( e j ω , n ) − ˆ ∠ G ( e j ω , n ) F ( e j ω , n ) = m 2 π , (phase cond.) ˆ F ( e j ω , n ) is the feedback path estimate. • The class of LMS algorithms break the magnitude condition • Freping breaks both magnitude and phase conditions 7/16
Freping for AFC and Frequency Warping in RT-MHA Feedback Path F ( z, n ) Freping y ( n ) Hearing Aid x ( n ) d ( n ) e ( n ) o ( n ) + Processing ? + + G ( z, n ) − B ( z ) A ( z, n ) y ( n ) ˆ u ( n ) Copy of W ( z, n ) d f ( n ) y f ( n ) ˆ u f ( n ) AFC Filter + + − A ( z, n ) W ( z, n ) e f ( n ) Coe ffi cient Adaptation 8
C-H Lee et al., Interspeech 2019 9 / 16
MPEG_es01_input.wav MPEG_es01_AFC.wav 10 / 16
MPEG_es01_FL.wav (frequency lowering in bands 4, 5) MPEG_es01_FU.wav (frequency increasing in bands 4, 5) 11 / 16
Machine Aided Self Fitting (Selfi) Research • Yeah, there’s an app for that! User preference (left) & Just Noticeable Differences • {skin, skim}, {state, skate}, {peer, poor}, {lock, locks}, … 12/16
The Machine’s Role in Selfi Research The structure of HA parameters is “known” – The structure of HA parameters is “unknown” – Closed form search techniques Stochastic search techniques 13/16
Big Data to Rescue NHANES (~30,000 PTAs) → Clustering → NAL-NL2 prescriptions → Binary Search Tree Fitting (BSTFit) → Selfi Refinement 14/16
How to enable psychophysical investigations beyond what is possible today? – OSP 1. What discoveries can clinical researchers make with the platform? 2. What discoveries can we translate to clinical practice? 15 / 16
Takeaway Message • Researchers from multiple disciplines – leverage contributions from others to advance their domain and • Participate in promoting hearing healthcare Further details Wednesday Morning (Crown) • – 3aPP3. Noise management features of the open speech platform – 3aPP4. Researcher and user interfaces for studies of hearing-aid self adjustment – 3aPP5. Open speech platform: Web-apps for hearing aids research • Wednesday Afternoon (Crown) – 3pSP15. Self-fit generation of the wide range compression parameters in hearing aids • http://openspeechplatform.ucsd.edu/ and https://github.com/nihospr01/OpenSpeechPlatform-UCSD 16 / 16
Backup 17
Collaborators (2018) 18
19 / 34
20
Recommend
More recommend
