8th Speech in Noise Workshop, 7-8 January 2016, Groningen

Speech perception studies with matrix sentence tests: A comparison across languages

Melanie Zokoll(a), Anna Warzybok(b), Sabine Hochmuth
Medizinische Physik and Cluster of Excellence „Hearing4all“, Carl von Ossietzky Universität Oldenburg, DE

Maria Boboshko
St. Petersburg Pavlov State Medical University, RU

Ruth Bentler
Hearing Aid Lab for Basic & Applied Research, University of Iowa, Iowa, US

Birger Kollmeier
Medizinische Physik and Cluster of Excellence „Hearing4all“, Carl von Ossietzky Universität Oldenburg and HörTech gGmbH, Oldenburg, DE

(a) Presenting
(b) Attending

It is known that testing a listener in a background of noise helps to assess information in addition to the pure-tone audiogram, such as possible supra-threshold distortions that occur in the auditory system as a result of the hearing impairment. By now, speech recognition tests in noise have become increasingly important in audiological diagnostics. Most of those tests use everyday sentences presented in an open-set format. So-called ‘matrix sentence tests’, however, use a closed-set format and comprise syntactically fixed, semantically unpredictable sentences (e.g. “Peter kept two green toys”) composed from a vocabulary of 50 words (10 alternatives for each word group). Matrix sentence tests are suitable for repeated speech perception testing in a multilingual society even when experimenters are not proficient in the test language (by using the closed-set response format). So far, matrix sentences tests are available for at least 15 different languages (e.g., in American English, French, German, Russian, Spanish, Turkish, and Italian) together with a varying degree of supportive data.

This contribution presents matrix sentence test data of multi-center studies in the USA, Canada and Russia investigating the reliability and comparability of the American English and Russian matrix sentence test, as well as the influence of the hearing ability on speech perception in quiet and noise. Data include adaptively estimated speech reception thresholds (SRTs), i.e. the sound pressure levels or signal-to-noise ratios (SNR) yielding 50% speech intelligibility, as well as correlations between hearing ability (pure tone average, PTA, for 0.5, 1, 2 and 4 kHz) and SRT. Results are that both tests provided reliable values with high comparability across languages. Individual SRTs in quiet correlated closely with audiograms. A comparatively poor relation was found between the SRT in quiet, or PTA and the SRT in noise. This supports the notion that the matrix sentence test in noise assesses an individual auditory factor that is separate from the audiogram and can help to disentangle the contribution of possible supra-threshold distortions to a certain hearing loss from that of the pure loss in sensitivity. Further, it can be concluded that matrix sentence tests are a sensitive diagnostic tool suitable for multilingual comparisons.

Last modified 2016-05-12 14:22:09