Algorithm development for signal enhancement

Modern digital hearing devices go far beyond simple sound-amplification devices. They may include extensive signal-processing algorithms adapted to the needs of the individual hearing impaired, and to the prevailing ambient acoustical conditions. This part of the HearCom projects concentrates on what types of algorithm are most effective for the individual hearing-impaired listener.

Adverse listening conditions

Apart from signal compression and frequency-dependent amplification tuned to the user's hearing profile, modern hearing aids and cochlear implants nowadays standardly incorporate several features to enhance the signal quality and listening effort in adverse listening conditions. Whereas mere compression and amplification might easily re-enable the hearing aid or cochlear implant user to maintain fairly normal conversation in relatively quiet environments, the intelligibility goes rapidly down in more challenging situations, such as environments with significant background noise or room reflections.

Signal enhancement techniques

During the past decades large numbers of signal processing schemes have been proposed to re-establish intelligibility in such adverse listening conditions, some of them with great, other with limited success. In the HearCom project five different, representative classes of signal enhancement techniques are considered :

  • Adaptive feedback cancellation. These are techniques to combat feedback using a neutralizing adaptive electronic feedback path incorporated in the device.
  • Adaptive beamforming. By combining signals from two or more microphones in the hearing aid or cochlear implant device, noise from a specific direction is suppressed, even if the direction of the source is changing.
  • Single-channel noise suppression. These algorithms require only one microphone, and attempt to separate interesting sounds from unwanted noise based on their statistical characteristics.
  • Blind source separation. This is a technique to separate sound sources, without knowing where they are. It is applied here using microphone signals from two hearing aid devices at both sides of the user's head.
  • Coherence-based dereverberation. This kind of approach suppresses diffuse noise and sound reflections from the room walls by combining signals from microphones from two hearing aid devices at both sides of the user's head.

Both state-of-the-art solutions that are currently available in the most advanced hearing aid and cochlear implant devices, as well as some promising novel techniques are studied in the project. Several algorithm variants belonging to the above mentioned classes have been implemented and evaluated. These variants differ in their theoretical approach and in the precise settings of signal-processing parameters that control their behaviour.

Sound demos

Sound demo material has been prepared for each of the considered algorithms, except for feedback cancellation, which is, strictly speaking, not a true signal enhancement approach. Sound files are available for the following subset of test conditions:

 
Algorithm
Room scenario Single-channel noise suppression Adaptive beamforming Blind source separation Coherence-based dereverberation
Low-reverberant room with stationary directional background noise unprocessed.mp3 processed.mp3 more information unprocessed.mp3 processed.mp3 more information unprocessed.mp3 processed.mp3 more information  
Living room with a directional music signal as background noise   unprocessed.mp3 processed.mp3 more information unprocessed.mp3 processed.mp3 more information  
Cafeteria with diffuse babble noise in the background unprocessed.mp3 processed.mp3 more information unprocessed.mp3 processed.mp3 more information   unprocessed.mp3 processed.mp3 more information