We sincerely appreciate the funding provided by the Oticon Foundation, the Natural Sciences and Engineering Research Council (NSERC), the Canada Foundation for Innovation (CFI), the Ontario Rehabilitation Technology Consortium, and the Health Technology Exchange of Ontario.
Objective evaluation of speech and audio quality
Objective assessment of speech and audio quality has several applications including evaluation of speech and audio coders, hearing aids, and assistive listening devices. Under this project, we are developing novel methods of objective estimation of speech quality which include an adaptive neuro-fuzzy technique which employs a neural network trained using perceptually relevant features and fuzzy logic rules and a statistical pattern recognition based speech quality estimator where the same perceptually relevant features were used to train Gaussian mixture density hidden Markov models (GMD-HMMs).
Impact of hearing aid signal processing technologies on speech and audio quality
Due to recent technological advances in digital signal processing (DSP) hardware and algorithms, a number of sophisticated DSP techniques are being developed and deployed in current generation hearing aids. These include new methods and algorithms for compression amplification, digital noise reduction, beamforming, binaural processing, and feedback cancellation. The goals of this project is to evaluate the impact of these algorithms on perceived speech and audio quality through instrumental and behavioural measures. Our aim is to develop a standardized measure for quantifying the speech quality which will allow clinical audiologists to assess the relative benefits of various devices that offer similar, but not identical, signal processing algorithms.
Adaptive modeling of hearing aid response to speech and music stimuli
This project applies adaptive modelling techniques to investigate the dynamic behavior of digital hearing aids when stimulated with natural speech and music signals. The adaptive modeling paradigm is very useful in estimating the distortion and noise in automatic signal processing hearing aids, which are not adequately measured using conventional electroacoustic procedures. We are particularly interested in developing subband adaptive filtering algorithms that model the dynamics of multi-channel compression hearing aids. Our earlier work has shown that subband adaptive models outperform full band adaptive models in accurately modeling the dynamic behavior of multi-channel compression hearing aids. However, the modeling performance of the subband adaptive structures was found to be suboptimal when there is a difference between the number of channels in the adaptive model and the hearing aid. We are currently investigating methods to identify the subband architecture of the hearing aid in order to build the proper subband adaptive model.
Speech coder quality and hearing impairment
The goal of this project is to evaluate the performance of speech coding algorithms that are currently used in wireless and internet voice communication systems. The speech coder performance is measured both objectively and subjectively. Objective measures include the speech intelligibility index (ANSI Standard), PESQ (International Telecommunications Union (ITU) standard), and our own speech quality metrics. Perceptual measurements of speech coder performance include the Hearing In Noise Test (HINT) for measuring the reception thresholds in noise, and speech quality ratings. The perceptual data will be collected from both normal hearing and hearing impaired listeners. Key research questions addressed in this study are:
- How do speech coding algorithms impact the speech intelligibility in noise? Is the impact compounded by hearing loss?
- What is the effect of speech coding on speech quality? Does the perception of speech quality differ between normal and hearing impaired listeners?
- What effect does "packet loss" have on speech intelligibility and quality, for normal and hearing impaired listeners? Do techniques aimed at alleviating the packet loss effect work well, especially with hearing impaired listeners?
- Can the perceptual measurements of speech intelligibility and quality be predicted using objective measures?