The multilingual matrix test: Principles, applications, and comparison across languages: A review

Abstract

Objective: A review of the development, evaluation, and application of the so-called ‘matrix sentence test’ for speech intelligibility testing in a multilingual society is provided. The format allows for repeated use with the same patient in her or his native language even if the experimenter does not understand the language. Design: Using a closed-set format, the syntactically fixed, semantically unpredictable sentences (e.g. ‘Peter bought eight white ships’) provide a vocabulary of 50 words (10 alternatives for each position in the sentence). The principles (i.e. construction, optimization, evaluation, and validation) for 14 different languages are reviewed. Studies of the influence of talker, language, noise, the training effect, open vs. closed conduct of the test, and the subjects’ language proficiency are reported and application examples are discussed. Results: The optimization principles result in a steep intelligibility function and a high homogeneity of the speech materials presented and test lists employed, yielding a high efficiency and excellent comparability across languages. The characteristics of speakers generally dominate the differences across languages. Conclusion: The matrix test format with the principles outlined here is recommended for producing efficient, reliable, and comparable speech reception thresholds across different languages.

Key Words::

• Speech intelligibility in noise
• speech reception threshold
• effect of language

Acknowledgements

This work has been supported by the CEC (Projects Natasha, HearCom), European regional funds (EFRE-Project HurDig), and by Deutsche Forschungsgemeinschaft (KO 942-13, EXC 1077 Hearing4all). The authors express their deep gratitude to all cooperation and project partners that contributed to the test implementations and validations in several languages. Special thanks goes to the HearCom project team, especially Tammo Houtgast, Marcel Vlaming, and Wouter Dreschler (who created the expression ‘matrix test’ as an international generalization of the OlSa-test). Mark Lutman, Lena Wong, Darrin Reed, and two anonymous reviewers are gratefully acknowledged for helping to improve the manuscript. The copyright of the different matrix tests is owned by the non-profit organization HörTech gGmbH (owned in majority by Universität Oldenburg) as well as Universities of Southampton, Rotterdam, Karolinska Institute Stockholm, Sør-Trøndelag University College, Norway, and other universities from the Hearcom consortium. An increasing number of tests are available as a medical product for modern audiometers of different brands (currently available for eight languages as a medical product). For research purposes, sample sentences and free trial versions of the research version of the software are also available from the copyright owners.

Notes

Declaration of interest: The authors report no conflicts of interest. The authors B.K. and V.U. are affiliated with the non-profit organization HörTech gGmbH which owns copyrights of the German matrix test and produces commercial, CE-certified software to perform the matrix test.

Notes

1. The original test by Hagerman did not include the co-articulation transitions, but segmented the words from 10 recorded sentences (uttered such that small pauses are created between each word) by cutting in the pauses between the words. For the Norwegian matrix test, a diphone-splitting method was used to cut the sentences into single words. Using this method, the cutting point was set in the midpoint of the first vowel in the succeeding word instead of splitting between the words (Øygarden, 2009).

2. Alternatively, the noise may be generated by filtering white noise to achieve the same frequency spectrum as the LTASS of the corpus of sentence materials, as has been performed for the British English version. However, such a procedure produces less speech-babble like temporal modulations that can be detected when comparing both types of noise via headphones (but not necessarily when presenting them in the free field). Hence, apart from the British English matrix test, the noise filtering procedure was not employed for the tests described here.

3. Note that Equation (1) assumes an infinite number of response alternatives and a random hit rate of 0 even though the number of response alternatives is limited to 10 words at each position. However, an additional response alternative (marked with a ‘?’) was provided in most of the tests to indicate that the subject had no definite response, in which case the response was counted as incorrect and the probability of guessing the correct word was 0. If this ‘null’ response alternative was not provided, the psychometric function had to be shifted by the random hit rate 1/A (where A is the number of response alternatives, i.e. A = 10 for the matrix test), and re-normalized as follows: (2)