Construction and evaluation of the Mandarin Chinese matrix (CMNmatrix) sentence test for the assessment of speech recognition in noise

Abstract

Objective: Development of the Mandarin Chinese matrix (CMNmatrix) sentence test for speech intelligibility measurements in noise according to the international standard procedure.

Design: A 50-word base matrix representing the distribution of phonemes and lexical tones of spoken Mandarin was established. Hundred sentences capturing all the co-articulations of two consecutive words were recorded. Word-specific speech recognition functions, speech reception thresholds (SRT: signal-to-noise ratio (SNR), that provides 50% speech intelligibility) and slopes were obtained from measurements at fixed SNRs. The speech material was homogenised in intelligibility by applying level corrections up to ± 2?dB. Subsequently, the CMNmatrix test was evaluated, the comparability of test lists was measured at two fixed SNRs. To investigate the training effect and establish the reference data, speech recognition was measured adaptively.

Study sample: Overall, the study sample contained 80 normal-hearing native Mandarin-speaking listeners.

Results: Multi-centre evaluation measurements confirmed that test lists are equivalent in intelligibility, with a mean SRT of ?10.1?±?0.1?dB SNR and a slope of 13.1?±?0.9 %/dB. The reference SRT is ?9.3?±?0.8 and ?11.2?±?1.2?dB SNR for the open- and closed-set response format, respectively.

Conclusion: The CMNmatrix test is suitable for accurate and internationally comparable speech recognition measurements in noise.     

An Italian matrix sentence test for the evaluation of speech intelligibility in noise

Objective: Development of an Italian matrix sentence test for the assessment of speech intelligibility in noise. Design: The development of the test included the selection, recording, optimization with level adjustment, and evaluation of speech material. The training effect was assessed adaptively during the evaluation measurements with six lists of 20 sentences, using open- and closed-set response formats. Reference data were established for normal-hearing listeners with adaptive measurements. Equivalence of the test lists was investigated using the open-set response format at three signal-to-noise ratios (SNRs). Study sample: A total of 55 normal-hearing Italian mother-tongue listeners. Results: The evaluation measurements at fixed SNRs resulted in a mean speech reception threshold (SRT) of ? 7.3 ± 0.2 dB SNR and slope of 13.3 ± 1.2 %/dB. The major training effect of 1.5 dB was observed for the first two consecutive measurements. Mean SRTs of ? 6.7 ± 0.7 dB SNR and ? 7.4 ± 0.7 dB SNR were found from the third to the sixth adaptive measurement for open- and closed-set test response formats, respectively. Conclusions: A good agreement has been found between the SRTs and slope and those of other matrix tests. Since sentences are difficult to memorize, the Italian matrix test is suitable for repeated measurements.     

Development and evaluation of the Turkish matrix sentence test

Objectives: The Turkish matrix sentence test, TURMatrix, was developed for precise, internationally comparable speech intelligibility testing. Design: The TURMatrix comprises a base matrix of ten well-known Turkish names, numbers, adjectives, objects, verbs, from which syntactically fixed sentences were randomly composed. Test conduction may be in an open-set (standard), or closed-set response format. Homogeneity in intelligibility of the test material was optimized by applying level adaptations (maximal ± 3 dB) based on word-specific speech reception thresholds (SRTs). Test list equivalence was verified and reference values were determined. Study sample: Thirty-eight native listeners of Turkish with normal hearing. Results: After training, mean SRT and slope of the final test lists were ? 8.3 ± 0.2 dB SNR and 14.1 ± 1.0%/dB, respectively (fixed SNR measurements; inter-list variability). For adaptive measurements, average across listeners was ? 7.2 ± 0.7 dB SNR in the open-set and ? 7.9 ± 0.7 dB SNR in the closed-set response format. Mean SRT for adaptive measurements in the open-set response format in quiet was 20.3 ± 4.1 dB. Individual SRTs in quiet correlated more closely with audiograms than with SRTs in noise. Conclusions: The TURMatrix was developed according to European standards and provides reliable speech intelligibility measurements in noise and quiet.     

The development and evaluation of the Finnish digit triplet test

Objectives: The aim of the study was to develop a reliable and easily accessible screening test for primary detection of hearing impairment.

Methods: Digits 0–9 were used to form quasirandom digit triplets. First, digit specific intelligibility functions and speech recognition thresholds (SRTs) were determined. To homogenize the test material digits with steep intelligibility function slopes were chosen and level correction up to ±2?dB were applied to the digits as needed. Evaluation measurements were performed to check for systematic differences in intelligibility between the test lists and to obtain normative reference function for normal-hearing listeners.

Results: The mean SRT and the final slope of the test lists were ?10.8?±?0.1?dB signal-to-noise ratio (SNR) and 21.7?±?1.8%/dB, respectively (measurements at constant level; inter-list variability). The mean SRT and slope of the test subjects were ?10.8?±?0.5?dB SNR and 23.4?±?5.2%/dB (measurements at constant level; inter-subject variability). The mean SRT for normal-hearing young adults for a single adaptive measurement is ?9.8?±?0.9?dB SNR.

Conclusion: The Finnish digit triplet test is the first self-screening hearing test in the Finnish language. It was developed according to current standards, and it provides reliable and internationally comparable speech intelligibility measurements.     

Polish sentence tests for measuring the intelligibility of speech in interfering noise

The aim of this study was to develop Polish sentence tests for accurate measuring of speech intelligibility in masking interfering noise. Two sets of sentence lists have been developed. The first set was composed of 25 lists and was used for sentence intelligibility scoring. The second set was composed of 22 lists and was used for word intelligibility scoring. The lists in each set have been phonemically and statistically balanced. The speech reception threshold (SRT) and slope of the psychometric function at the SRT point (S₅₀) were determined in normal-hearing subjects. It was found that the mean SRT and mean list-specific S_50list for the first set were equal to ?6.1 dB and 25.5%/dB, respectively. The mean SRT and the mean list-specific S_50list for the second set were:?7.5 dB and 20.8%/dB. Due to a relatively steep slope of the psychometric functions, the Polish sentence tests were shown to be accurate materials for speech intelligibility measurements against interfering noise. They are the first sentence speech-in-noise tests developed for Slavonic languages.     

Characteristics and international comparability of the Finnish matrix sentence test in cochlear implant recipients

Objectives: The first Finnish sentence-based speech test in noise—the Finnish matrix sentence test—was recently developed. The aim of this study was to determine the characteristics of the new test with respect to test-retest reliability, speech recognition curve, and international comparability in Finnish cochlear implant (CI) recipients. Design: The speech reception thresholds (SRT) were measured by means of an adaptive test procedure and compared with the results of the traditional Finnish word test. Additional measurements for concurrent slope and SRT estimation were conducted to determine the speech recognition curve and to check the test-retest reliability. Study sample: The measurements were performed on 78 Finnish CI recipients. In a subset of 25 patients, additional measurements for test-retest reliability and slope determination were performed. Results: The mean SRT was ? 3.5 ± 1.7 dB SNR, with only a weak correlation with the Finnish word test. Test-retest reliability was within ± 1 dB and the mean slope of the speech recognition curve was 14.6 ± 3.6 %/dB. The rehabilitation results were similar to the results published for the German matrix test. Conclusions: The Finnish matrix test was found to be suitable and efficient in CI recipients with similar characteristics as the German matrix test.     

Do you hear the noise? The German matrix sentence test with a fixed noise level in subjects with normal hearing and hearing impairment

Objective: The aim of this study was to determine the relationship between hearing loss and speech reception threshold (SRT) in a fixed noise condition using the German Oldenburg sentence test (OLSA). Design: After training with two easily-audible lists of the OLSA, SRTs were determined monaurally with headphones at a fixed noise level of 65 dB SPL using a standard adaptive procedure, converging to 50% speech intelligibility. Study sample: Data was obtained from 315 ears of 177 subjects with hearing losses ranging from ? 5 to 90 dB HL pure-tone average (PTA, 0.5, 1, 2, 3 kHz). Results: Two domains were identified with a linear dependence of SRT on PTA. The SRT increased with a slope of 0.094 ± 0.006 dB SNR/dB HL (standard deviation (SD) of residuals = 1.17 dB) for PTAs < 47 dB HL and with a slope of 0.811 ± 0.049 dB SNR/dB HL (SD of residuals = 5.54 dB) for higher PTAs. Conclusion: The OLSA can be applied to subjects with a wide range of hearing losses. With 65 dB SPL fixed noise presentation level the SRT is determined by listening in noise for PTAs < ～47 dB HL, and above it is determined by listening in quiet.     

Design,optimization and evaluation of a Danish sentence test in noise

The Danish sentence test DANTALE II was developed in analogy to the Swedish sentence test by Hagerman and the German Oldenburg sentence test as a new Danish sentence test in noise to determine the speech reception threshold in noise (SRT, i.e. the signal-to-noise ratio (SNR) that yields 500% intelligibility). Each sentence is generated by a random combination of the alternatives of a base list. This base list consists of 10 sentences with the same syntactical structure (name, verb, numeral, adjective, object). The test sentences were recorded and segmented in such a way that the coarticulation effects were taken into account in order to achieve a high perceived sound quality of the resynthesized sentences: 100 sentences were recorded, each coarticulation between each word and the 10 possible following word alternatives were recorded, and the correct coarticulation was used to generate the test sentences. Word-specific speech recognition curves were measured for each recorded word to optimize the homogeneity of the speech material and the measurement accuracy. Level corrections of particular words and a careful selection of the test lists produced a noticeable reduction in the variation in the distribution of word-specific SRT (standard deviation 1.75 dB instead of 3.78 dB). Therefore, the slope of the total intelligibility function was expected to increase from 8.30%/dB (raw test material) to 13.2%/dB (after modification). These theoretical expectations were evaluated by independent measurements with normal-hearing subjects, and, for the most part, confirmed. The reference data for the DANTALE II are: SRT=-8.43 dB SNR; slope at SRT, s50 = 13.2%/dB. The training effect was 2.2 dB and could be reduced to less than 1 dB if two training lists of 20 sentences were performed prior to data collection.     

A Spanish matrix sentence test for assessing speech reception thresholds in noise

Objective: To develop, optimize, and evaluate a new Spanish sentence test in noise. Design: The test comprises a basic matrix of ten names, verbs, numerals, nouns, and adjectives. From this matrix, test lists of ten sentences with an equal syntactical structure can be formed at random, with each list containing the whole speech material. The speech material represents the phoneme distribution of the Spanish language. The test was optimized for measuring speech reception thresholds (SRTs) in noise by adjusting the presentation levels of the individual words. Subsequently, the test was evaluated by independent measurements investigating the training effects, the comparability of test lists, open-set vs. closed-set test format, and performance of listeners of different Spanish varieties. Study sample: In total, 68 normal-hearing native Spanish-speaking listeners. Results: SRTs measured using an adaptive procedure were 6.2 ± 0.8 dB SNR for the open-set and 7.2 ± 0.7 dB SNR for the closed-set test format. The residual training effect was less than 1 dB after using two double-lists before data collection. Conclusions: No significant differences were found for listeners of different Spanish varieties indicating that the test is applicable to Spanish as well as Latin American listeners. Test lists can be used interchangeably.     

Influence of noise type on speech reception thresholds across four languages measured with matrix sentence tests

Objective: To compare speech reception thresholds (SRTs) in noise using matrix sentence tests in four languages: German, Spanish, Russian, Polish. Design: The four tests were composed of equivalent five-word sentences and were all designed and optimized using the same principles. Six stationary speech-shaped noises and three non-stationary noises were used as maskers. Study sample: Forty native listeners with normal hearing: 10 for each language. Results: SRTs were about 3 dB higher for the German and Spanish tests than for the Russian and Polish tests when stationary noise was used that matched the long-term frequency spectrum of the respective speech test materials. This general SRT difference was also observed for the other stationary noises. The within-test variability across noise conditions differed between languages. About 56% of the observed variance was predicted by the speech intelligibility index. The observed SRT benefit in fluctuating noise was similar for all tests, with a slightly smaller benefit for the Spanish test. Conclusions: Of the stationary noises employed, noise with the same spectrum as the speech yielded the best masking. SRT differences across languages and noises could be attributed in part to spectral differences. These findings provide the feasibility and limits of comparing audiological results across languages.     

Internationally comparable screening tests for listening in noise in several European languages: The German digit triplet test as an optimization prototype

Abstract

Objective: A review is given of internationally comparable speech-in-noise tests for hearing screening purposes that were part of the European HearCom project. This report describes the development, optimization, and evaluation of such tests for headphone and telephone presentation, using the example of the German digit triplet test. In order to achieve the highest possible comparability, language- and speaker-dependent factors in speech intelligibility should be compensated for. Materials and methods: The tests comprise spoken numbers in background noise and estimate the speech reception threshold (SRT), i.e. the signal-to-noise ratio (SNR) yielding 50% speech intelligibility. Results: The respective reference speech intelligibility functions for headphone and telephone presentation of the German version for 15 and 10 normal-hearing listeners are described by a SRT of ?9.3?±?0.2 and ?6.5?±?0.4 dB SNR, and slopes of 19.6 and 17.9%/dB, respectively. Reference speech intelligibility functions of all digit triplet tests optimized within the HearCom project allow for investigation of the comparability due to language specificities. Conclusions: The optimization criteria established here should be used for similar screening tests in other languages.     

Development of a Dutch matrix sentence test to assess speech intelligibility in noise

Abstract

Objective: A Dutch matrix sentence test was developed and evaluated. A matrix test is a speech-in-noise test based on a closed speech corpus of sentences derived from words from fixed categories. An example is “Mark gives five large flowers.” Design: This report consists of the development of the speech test and a multi-center evaluation. Study sample: Forty-five normal-hearing participants. Results: The developed matrix test has a speech reception threshold in stationary noise of ? 8.4 dB with an inter-list standard deviation of 0.2 dB. The slope of the intelligibility function is 10.2 %/dB and this is slightly lower than that of similar tests in other languages (12.6 to 17.1 %/dB). Conclusions: The matrix test is now also available in Dutch and can be used in both Flanders and the Netherlands.     

Development of equally intelligible Telugu sentence-lists to test speech recognition in noise

Objective: To develop sentence lists in the Telugu language for the assessment of speech recognition threshold (SRT) in the presence of background noise through identification of the mean signal-to-noise ratio required to attain a 50% sentence recognition score (SRTn). Design: This study was conducted in three phases. The first phase involved the selection and recording of Telugu sentences. In the second phase, 20 lists, each consisting of 10 sentences with equal intelligibility, were formulated using a numerical optimisation procedure. In the third phase, the SRTn of the developed lists was estimated using adaptive procedures on individuals with normal hearing. Study sample: A total of 68 native Telugu speakers with normal hearing participated in the study. Of these, 18 (including the speakers) performed on various subjective measures in first phase, 20 performed on sentence/word recognition in noise for second phase and 30 participated in the list equivalency procedures in third phase. Results: In all, 15 lists of comparable difficulty were formulated as test material. The mean SRTn across these lists corresponded to ?2.74 (SD?=?0.21). Conclusions: The developed sentence lists provided a valid and reliable tool to measure SRTn in Telugu native speakers.     

The effect of fluctuating maskers on speech understanding of high-performing cochlear implant users

Objective: The present study evaluated whether the poorer baseline performance of cochlear implant (CI) users or the technical and/or physiological properties of CI stimulation are responsible for the absence of masking release. Design: This study measured speech reception thresholds (SRTs) in continuous and modulated noise as a function of signal to noise ratio (SNR). Study sample: A total of 24 subjects participated: 12 normal-hearing (NH) listeners and 12 subjects provided with recent MED-EL CI systems. Results: The mean SRT of CI users in continuous noise was??3.0?±?1.5 dB SNR (mean?±?SEM), while the normal-hearing group reached??5.9?±?0.8 dB SNR. In modulated noise, the difference across groups increased considerably. For CI users, the mean SRT worsened to??1.4?±?2.3 dB SNR, while it improved for normal-hearing listeners to??18.9?±?3.8 dB SNR. Conclusions: The detrimental effect of fluctuating maskers on SRTs in CI users shown by prior studies was confirmed by the current study. Concluding, the absence of masking release is mainly caused by the technical and/or physiological properties of CI stimulation, not just the poorer baseline performance of many CI users compared to normal-hearing subjects. Speech understanding in modulated noise was more robust in CI users who had a relatively large electrical dynamic range.     

Talker- and language-specific effects on speech intelligibility in noise assessed with bilingual talkers: Which language is more robust against noise and reverberation?

Objective: Investigate talker- and language-specific aspects of speech intelligibility in noise and reverberation using highly comparable matrix sentence tests across languages. Design: Matrix sentences spoken by German/Russian and German/Spanish bilingual talkers were recorded. These sentences were used to measure speech reception thresholds (SRTs) with native listeners in the respective languages in different listening conditions (stationary and fluctuating noise, multi-talker babble, reverberated speech-in-noise condition). Study sample: Four German/Russian and four German/Spanish bilingual talkers; 20 native German-speaking, 10 native Russian-speaking, and 10 native Spanish-speaking listeners. Results: Across-talker SRT differences of up to 6 dB were found for both groups of bilinguals. SRTs of German/Russian bilingual talkers were the same in both languages. SRTs of German/Spanish bilingual talkers were higher when they talked in Spanish than when they talked in German. The benefit from listening in the gaps was similar across all languages. The detrimental effect of reverberation was larger for Spanish than for German and Russian. Conclusions: Within the limitations set by the number and slight accentedness of talkers and other possible confounding factors, talker- and test-condition-dependent differences were isolated from the language effect: Russian and German exhibited similar intelligibility in noise and reverberation, whereas Spanish was more impaired in these situations.     

The French digit triplet test: A hearing screening tool for speech intelligibility in noise

Abstract

A French speech intelligibility screening test in noise that applies digit triplets as stimuli has been developed and evaluated for both telephone and broadband headphone use. After optimizing the speech material based on the intelligibility of the individual digits, norms for normal-hearing subjects were established. speech reception thresholds (SRTs) of ?6.4 ± 0.4 and ?10.5 ± 0.3 dB SNR, and slopes of 17.1 and 27.1 %/dB were obtained for telephone and broadband headphone presentation, respectively. The French digit triplet test by telephone was then implemented as an automatic self-screening test by home telephone, and further evaluated in normal-hearing and hearing-impaired listeners. A test-retest variability of 0.7 dB was found and the correlation between SRT and pure-tone average (PTA_0.5,1,2,4) was 0.77. One month after launching the test, 20 000 calls were registered. It can be concluded that both versions of the newly developed test have steep slopes and small SRT differences across normal-hearing listeners. The screening test by telephone is highly reliable and proves to fulfill the need for an easily accessible and objective hearing screening.

Sumario

Se desarrolló y evaluó una prueba de tamiz de inteligibilidad del lenguaje en ruido, utilizando ternas de dígitos como estímulo, tanto para teléfono como para auriculares de banda ancha. Después de optimizar el material lingüístico basado en la inteligibilidad de los dígitos individuales, se establecieron las normas para los sujetos normoyentes. Se obtuvieron umbrales de recepción del lenguaje (SRTs) de -6.4 ± 0.4 y -10.5 ± 0.3dB SNR y pendientes de 17.1 y 27.1%/dB para teléfono y auriculares de banda ancha, respectivamente. Se implementó la prueba de terna de dígitos en francés por teléfono como una prueba de tamiz autoaplicable por teléfono en casa y después se evaluó en normoyentes e hipoacúsicos. Se encontró una variabilidad de test-re-test de 0.7 dB y la correlación entre SRT y el promedio de tonos puros (PTA_0.5,1,2,4) fue de 0.77. Un mes después de lanzar la prueba, se registraron 20,000 llamadas. Se puede concluir que ambas versiones de la prueba recientemente desarrollada tiene pendientes pronunciadas y pequeñas diferencias del SRT entre los normoyentes. La prueba de tamiz por teléfono es altamente confiable y demuestra que satisface las necesidades de una prueba de tamiz auditivo fácil, accesible y objetiva.     

Speech intelligibility of virtual humans

Abstract

The speech intelligibility benefit of visual speech cues during oral communication is well-established. Therefore, an ecologically valid approach of auditory assessment should include the processing of both auditory and visual speech cues. This study describes the development and evaluation of a virtual human speaker designed to present speech auditory-visually. A male and female virtual human speaker were created and evaluated in two experiments: a visual-only speech reading test of words and sentences and an auditory-visual speech intelligibility sentence test. A group of five hearing, skilled speech reading adults participated in the speech reading test whereas a group of young normal-hearing participants (N?=?35) was recruited for the intelligibility test. Skilled speech readers correctly identified 57 to 67% of the words and sentences uttered by the virtual speakers. The presence of the virtual speaker improved the speech intelligibility of sentences in noise by 1.5 to 2?dB. These results demonstrate the potential applicability of virtual humans in future auditory-visual speech assessment paradigms.     

Development and evaluation of a mixed gender,multi-talker matrix sentence test in Australian English

Objective: To develop, in Australian English, the first mixed-gender, multi-talker matrix sentence test. Design: Speech material consisted of a 50-word base matrix whose elements can be combined to form sentences of identical syntax but unpredictable content. Ten voices (five female and five male) were recorded for editing and preliminary level equalization. Elements were presented as single-talker sentences-in-noise during two perceptual tests: an optimization phase that provided the basis for further level correction, and an evaluation phase that perceptually validated those changes. Study sample: Ten listeners participated in the optimization phase; these and an additional 32?naïve listeners completed the evaluation test. All were fluent in English and all but one had lived in Australia for >2 years. Results: Optimization reduced the standard deviation (SD) and speech reception threshold (SRT) range across all speech material (grand mean SRT?=??10.6?dB signal-to-noise ratio, median?=??10.8, SD =1.4, range =13.7, slope =19.3%/dB), yielding data consistent with cross-validated matrix tests in other languages. Intelligibility differences between experienced and naïve listeners were minimal. Conclusions: The Australian matrix corpus provides a robust set of test materials suitable for both clinical assessment and research into the dynamics of active listening in multi-talker environments.     

Adaptation of the STARR test for adult Italian population: A speech test for a realistic estimate in real-life listening conditions

Objectives: To introduce the Italian adaptation of the STARR test based on a roving-level adaptive method to mimic challenging real-life listening conditions for use in people with auditory prostheses. Design: Normative data were collected and interlist-variability, as well as learning effects, were investigated using a within-subject design with repeated measures. Study sample: A group of 32 normal-hearing (NH) adults participated in the study. Results: The average speech reception threshold (SRT) for NH subjects was??8.4 dB SNR. The variability of mean SRTs across test lists was relatively small (≤1 dB for all test lists). The statistically significant differences between lists were eliminated after applying correction factors. On the basis of variability for the corrected SRTs within each subject, a difference of 2.8 dB in SRT was meaningful for outcome comparisons using one test list per condition and 2 dB using two lists per condition. Statistical analysis did not show any significant learning effects. Conclusions: Findings in NH listeners suggested that the Italian STARR test could be a promising supplement to existing speech assessment tools. Further studies in populations with hearing impairment could contribute to cross-language studies.     

Adaptation of the HINT (hearing in noise test) for adult Canadian Francophone populations

The HINT provides an efficient and reliable method of assessing speech intelligibility in quiet and in noise by using an adaptive strategy to measure speech reception thresholds for sentences, thus avoiding ceiling and floor effects that plague traditional measures performed at fixed presentation levels. A strong need for such a test within the Canadian Francophone population, led us to develop a French version of the HINT. Here we describe the development of this test. The Canadian French version is composed of 240-recorded sentences, equated for intelligibility, and cast into 12 phonemically balanced 20-sentence lists. Average headphone SRTs, measured with 36 adult Canadian Francophone native speakers with normal hearing, were 16.4?dBA in quiet, ?3.0?dBA SNR in a 65?dBA noise front condition and ?11.4?dBA SNR in a 65?dBA noise side condition. Reliability was established by means of within-subjects standard deviation of repeated SRT measurements over different lists and yielded values of 2.2 and 1.1?dB for the quiet and noise conditions, respectively.