Comparison of predictive measures of speech recognition after noise reduction processing
In hearing aid research there is a constant need for the development of verification measures that could reliably predict and thus, at least partially, replace often time-consuming listening tests performed by test subjects. Hence, the aim of this study was to evaluate a number of objective measures concerning their qualification for predicting perceived sound quality and experimentally measured speech-recognition benefit of noise reduction algorithms.
Speech test material was presented to 20 hearing impaired and 10 normal hearing listeners, both in a reference condition and processed by three NR algorithms, respectively. An adaptive procedure was used to obtain individual, NR algorithm related signal-to-noise ratios which correspond to equal speech-recognition performance. This allowed for a direct comparison between the measured and predicted NR algorithm performance. In addition, the results of sound quality ratings were compared to predicted sound quality. Only one out of nine tested predictive measures could predict the speech test results reliably for both listener groups. In general, prediction performance was better for the group of hearing impaired listeners and the majority of correctly predicting measures performed short-time analyses and based their calculations on the correlation between clean speech and the processed noisy speech. The results from the sound quality related measures were difficult to interpret, but most of the predictive measures showed the trend to rank order the three noise reduction algorithms correctly.
Publications and presentations
Smeds K, Leijon A, Wolters F, Hammarstedt A, Båsjö S, Hertzman S. (2014) Comparison of predictive measures of speech recognition after noise reduction processing. Journal of Acoustical Society of America 136: 1363-1374. Nilsson, A. (2010). Evaluation of theoretical measures to assess their ability to predict the performance of noise reduction algorithms. (MSc), KTH, Stockholm.
Wolters, F. (2010). Predictive ability of objective methods in terms of speech intelligibility and sound quality for noise reduction systems in hearing aids. (Bachelor of Engineering), University of Applied Sciences, Oldenburg, Stockholm.
ISAAR (International Symposium on Auditory and Audiological Research) in Nyborg (Denmark), August 2011
STAF (National Swedish audiology conference for engineers and technicians), Uppsala (Sweden), 2011.
AES (International conference of the Audio Engineering Society), Piteå (Sweden), 2010.
Audionomdagarna (National Swedish conference for Audiologists), Västerås (Sweden), 2010.
STAF (National Swedish conference for engineers and technicians), Eskilstuna (Sweden), 2010.
IHCON (International Hearing Aid Research Conference), Lake Tahoe, California (USA), 2010.