Technology Update SessionSession 3A Use of Artificial Intelligence for denoising has been available in consumer technology for years. These systems often rely on deep-learning approaches to identify and remove noise from the audio stream, which improves the resulting signal-to-noise ratio (SNR) of the processed audio. A deep-learning denoising system applied in a hearing aid should similarly be capable of improving short-term SNR while enhancing target speech, in contrast to conventional single microphone noise reduction that improves listening comfort while improving long-term SNR. Despite these clear advantages, hearing aids have historically not had the technical capability to operate such computationally intensive algorithms. Phonak Audéo Sphere Infinio utilizes two processors, one 'traditional' chipset that drives core functions of the hearing aid, and a second purpose-built chipset called DEEPSONIC that operates a deep-neural-network (DNN) denoising system. The processing power and efficiency of DEEPSONIC allows the DNN denoising system to estimate the real and complex components of the environmental noise, enabling the device to effectively remove noise while preserving speech. In a clinical trial of Audéo Sphere Infinio, speech intelligibility and listening effort were measured for 27 adults with moderate-to-severe hearing loss at the Phonak Audiology Research Center (PARC). During the speech intelligibility task, sentences were presented from one of four off-axis locations while speech-shaped noise was presented constantly from all loudspeakers at an SNR of -3 dB. Presentation location was randomized across sentences. When controlling for all other sound cleaning features, activation of DNN denoising doubled the odds of correctly repeating a sentence. The Adaptive Categorical Listening Effort Scaling (ACALES) task was used to assess listening effort for speech presented at 0 degrees azimuth in diffuse multi-talker babble noise. When controlling for the effects of all other sound cleaning features, activating DNN denoising allowed participants to tolerate a 2.9 dB lower (i.e., poorer) SNR without a corresponding increase in perceived effort. The results of this clinical trial demonstrate that DNN denoising, enabled by the DEEPSONIC chipset and available in Audéo Sphere Infinio, expands upon the existing benefit of legacy technology noise reduction for listening comfort by improving outcomes in spatially complex listening environments like those encountered during daily life.
|