Clearer Conversations: Reinventing Hearing Aid Stability

Have you ever been in a lecture hall or at a concert and heard a high-pitched screeching sound from the speakers? It happens when a microphone gets very close to a loudspeaker. It picks up the sound from the speaker and amplifies it, forming a loop until it becomes a piercing or howling sound. This phenomenon is known as acoustic feedback. In hearing aids, which boost sound and speech while reducing the noise, this feedback limits the extent of amplification that can be safely provided to the user. Some common causes of users with hearing aids experiencing acoustic feedback include improper device fit, excessive earwax, high volume, and objects such as mobile phones directing sound back into the hearing aid microphone. 

Adaptive feedback cancellation attempts to effectively tackle this issue of acoustic feedback. In hearing aids, this involves the use of specific filters that ‘listen’ to the sound from the speaker. Upon listening, the filters produce a mirror-inverted sound of the high-pitched noise. When the microphone receives the high-pitched noise with the mirror-inverted sound, they cancel each other out. Simply put, the hearing aid predicts the feedback signal and subtracts it before it converts into a howl.  

For a long time, feedback cancellation systems assumed that the path between the speaker and the microphone behaved in a linear manner, meaning doubling the input would double the output. However, this does not hold for many real-life situations. Higher volumes start causing distortions in this linearity. A relatable example is that of an old phone speaker. Initially, the sound increases smoothly with its quality largely intact upon increasing the volume. But after a certain level, an increase in the volume causes the sound to become loud, harsh and distorted. Traditional feedback cancellation methods face problems due to this non-linearity, such as compromised feedback suppression and sound quality. Noisy environments, such as traffic congestion and crowded rooms and markets, magnify this issue. 

As an effort to address this problem, researchers from the Indian Institute of Technology Gandhinagar (IITGN) proposed an Enhanced Hammerstein-Spline Adaptive Filter (EHSAF). Their study was recently published in Signal Processing. The EHSAF is an improved version of the conventional Hammerstein-spline model. In simple words, this model reduces distortions, delays, and echoes, and makes the sound smooth, flexible, and natural by adapting to the user’s changing environment, such as from a quiet place to a noisy party. 

The model’s update rule can be seen as a “learning” process that adjusts it to be more accurate. The enhancement in EHSAF is with respect to this rule. “The EHSAF is an upgrade over the conventional model, which spends a lot of computational effort trying to model the entire path instead of focusing on the few critical points,” said Tarun Meena, a senior undergraduate in Electrical Engineering at IITGN. Instead of only having a limited focus on the present moment, the EHSAF also considers the influence of recent past experiences. This comprehensiveness allows the algorithm to remain stable even when the feedback path is sparse. The latter means that only some points of the feedback signal are crucial.   

Instead of treating the feedback path as a long and dense nonlinear filter, the researchers employed a mathematical technique called the nearest Kronecker product (NKP). “A low-rank NKP approximation retains only the dominant components that capture most of the energy or structure of the original system. The number of adaptive parameters is significantly decreased. This process lowers computational complexity because now we have features like structured operations that involve smaller factors and reduced memory requirements due to compact representation. These are significant enhancements over the existing nonlinear adaptive feedback cancellation methods,” explained Dr Shouharda Ghosh. Dr Ghosh is the first author of this study. He recently defended his PhD thesis in Electrical Engineering at IITGN, and this work was a part of his thesis. 

According to Dr Nithin V. George, “Importantly, the proposed framework enhances voice quality, intelligibility, and stability. By reducing the calculations required, the method yields faster and accurate results. In simple words, it would mean that the users can experience louder and clearer sounds without triggering the screeching noise. This framework can make conversations in noisy environments far more comfortable.” Dr George is a TEOCO Chair Professor in the Department of Electrical Engineering and the Principal Investigator at the Audio Signal Processing lab at IITGN.  

The number of elderly, aged 60 and above, in India is increasing day by day, comprising 12% of the country’s population according to the Longitudinal Ageing Study of India in 2021. Age-related hearing loss, being one of the most neglected and underdiagnosed issues in the country, further underscores the value of this research. Interestingly, the findings of this study may also be relevant beyond hearing aids. Devices such as tactical headsets used in defence environments and pilot communication headsets also face issues like sound distortions, echoes, and diverse acoustic conditions. Hence, the proposed framework can be beneficial in demanding situations where there is a strict requirement for audio systems to operate reliably.     

Since microphones can also exhibit nonlinear behaviour at higher sound levels, research focused on not only improving them, but also a combination of loudspeaker and microphone paths is a future direction. Another avenue of research is to incorporate data-driven approaches for analysing feedback data from different users in diverse environments. It would aid the designing of adaptive and robust models to suit individual differences. Further, the connection of hearing aids with phones and other such devices may make the system prone to hardware-level risks. These may range from accidentally revealing information through indirect signals (side-channel leakage) to deliberately disturbing the device and learning from its behaviours (fault-based attacks). Thus, future designs need to balance performance, energy efficiency, and system-level security for improved functioning of hearing aids. 

The researchers acknowledged the support of the Prime Minister’s Research Fellowship, Department of Science and Technology, Government of India, under the Mathematical Research Impact Centric Support (MATRICS) scheme and the Employee Owned Corporation Chair (TEOCO) Chair of IITGN.