The long term goal of this research is to create high-performance hearing aids to hear speech in acoustically cluttered environments. Difficulty in understanding speech is a primary cause of dissatisfaction with hearing aid technology for the majority of hearing aid users who are 65 years and older. The immediate goal is to develop two novel signal processing algorithms that process binaural inputs, optimally reject interfering sounds, and reconstruct the desired signal in real- time. One, inspired by our understanding of how biological auditory systems operate, localizes the interfering sound sources and steers a null in the reception pattern toward the largest in each frequency band. The second algorithm employs a variation of a technique known as minimum variance beamforming that was developed for radar and other engineering problems. It minimizes the total off-axis sound energy on a frequency by frequency basis. The novelty and power of the algorithms derives from the fact that they perform their operations on each of up to thousands of frequency channels simultaneously while adapting to changing acoustic signals at each time instant. In the biologically inspired algorithm the adaptation is driven by a novel technique for localizing multiple sound sources with only two microphones; in the minimum variance technique a new computational strategy makes possible real-time adaptation. Preliminary simulation and field testing indicates that both algorithms provide substantial improvement in signal-to-noise ratio and speech intelligibility. A first generation real-time system for the second algorithm has been successfully constructed, allowing us to evaluate its real-world performance and to efficiently identify and correct practical problems.
The specific aims of this two-year initiative are:
Aim number 1. To optimize the real-time implementation of one of our algorithms (the minimum variance beamformer) and to implement and optimize the other (the biologically inspired algorithm.).
Aim number 2. To modify the algorithms to account for changes in sensitivity due to positions of the microphones, especially by incorporating head-related transfer functions.
Aim number 3. To assess the performance of the algorithms computationally and to evaluate performance empirically for groups of older and younger listeners with and without hearing loss. The research approach is highly interdisciplinary, as is the research team (acoustician, electrical engineer, signal processing engineer, neurobiologist, audiologist, psychoacoustician).
| Larsen, Erik; Iyer, Nandini; Lansing, Charissa R et al. (2008) On the minimum audible difference in direct-to-reverberant energy ratio. J Acoust Soc Am 124:450-61 |
| Lockwood, Michael E; Jones, Douglas L; Bilger, Robert C et al. (2004) Performance of time- and frequency-domain binaural beamformers based on recorded signals from real rooms. J Acoust Soc Am 115:379-91 |
| Ratnam, Rama; Jones, Douglas L; Wheeler, Bruce C et al. (2003) Blind estimation of reverberation time. J Acoust Soc Am 114:2877-92 |
