Improving localization in binaural beamforming for hearing aid wearers
Hearing impairment often goes along with diminished frequency selectivity which can lead to speech understanding problems in noisy environments. In hearing aids this is typically addressed by using an advanced binaural beamforming algorithm to effectively attenuate background noise as well as non-target interfering speakers in “cocktail party scenarios” (Kamkar-Parsi et al., 2014). This improves speech understanding and decreases listening effort (e.g. Mejia et al., 2017; Bernarding et al., 2014). In specific laboratory situations, it can even result in better speech understanding of hearing impaired subjects compared to normal hearing unaided subjects (Fröhlich et al., 2015).
Another important aspect of binaural hearing is the ability to localize sound in order to be aware of one’s surrounding and speakers’ location. However, this is more difficult to maintain in binaural beamforming than in the omnidirectional mode. Binaural beamforming improves the Signal-To-Noise Ratio (SNR), but it also distorts spatial binaural cues to some extent. Here, we report on a study investigating a new algorithm (implemented in the Signia Nx platform) that improves the preservation of spatial cues by restoring phase information in the low frequencies while maintaining target speech understanding.
In a laboratory test with 19 hearing impaired subjects we compared Speech Reception Threshold (SRT) and localization ability for hearing aids with three algorithms: TrueEar program (i.e. mimicking the natural directivity of head and pinna), the beamforming algorithm used in a predecessor implementation and the improved beamforming algorithm (Signia Nx) with better preservation of spatial cues. Subjects were on average 69.6 years old (SD: 15.6 years) with a mean pure tone average of 54.3 dB. Hearing aids were individually fitted for each subject.
The localization performance was best in TrueEar mode. When using the binaural beamforming with phase cue restoration (Signia Nx implementation), localization performance was significantly improved compared to the predecessor implementation. Most importantly, a speech understanding in noise test showed no significant difference between the two beamforming modes.
It is also notable that subjects seem to adapt their localization strategy to the beamforming mode, i.e., their performance improved with time in the predecessor beamformer and decreased in True Ear. Such an adaptation effect was not observed for the enhanced (Signia Nx) beamformer.
This could indicate that in real life where hearing aids change between TruEar and beamforming modes, it is beneficial to have similar localization cues in both modes so that less adaptation is necessary and overall localization performance remains more stable.