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

Abstract

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.

• Speech intelligibility
• Sentence test
• Psychometric function
• Speech reception threshold

Abbreviations
CCITT	=	International Telegraph and Telephone Consultative Committee
SNR	=	Signal-to-noise ratio
SRT	=	Speech reception threshold

Abbreviations
CCITT	=	International Telegraph and Telephone Consultative Committee
SNR	=	Signal-to-noise ratio
SRT	=	Speech reception threshold

Notes

Sumario

El objetivo de este estudio fue desarrollar pruebas de frases en Polaco para medir con exactitud la inteligibilidad del lenguaje en medio de ruido de balbuceo enmascarante. Dos grupos de listas de frases han sido desarrolladas. El primer grupo se compuso de 25 listas y se uso para puntaje de inteligibilidad de frases. El segundo grupo se compuso de 22 listas y se uso para puntaje de inteligibilidad de palabras. Las listas en cada grupo fueron balanceadas fonémica y estadísticamente. El umbral de recepción del lenguaje (SRT) y la pendiente de la función psicométrica en el punto del SRT (S₅₀) se determinaron en individuos normo-oyentes. Se encontró que el SRT medio y la lista específica S_50list media para el primer grupo eran iguales a −6.1 dB y 25.5% dB, respectivamente. El SRT medio y la lista específica S_50list media para el segundo grupo fueron: −7.5 dB y 20.8% dB. Debido a la pendiente relativamente empinada de las funciones psicométricas, se vio que las pruebas de frases en Polaco eran materiales exactos para medidas de inteligibilidad del lenguaje contra el ruido de interferencia. Ellas constituyen la primera prueba de frases en ruido desarrolladas para las lenguas eslávicas.

Notes

1. Characterizing some statistical properties of the babble noise used in this study it was found that the crest factor C_f, reflecting the dynamic range of amplitude fluctuations, and normalized kurtosis (kurt), characterizing a relation of the distribution of instantaneous values to the normal distribution, were C_f=4.51 and kurt = 1.02 (kurt = µ₄/(3σ₄), where µ₄ is the fourth central moment and σ is the standard deviation). These values are similar to C_f=4.65 and kurt = 1.04, determined for masker used by Versfeld et al (2000), and to C_f=4.55 and kurt = 0.98, determined for masker used by Kollmeier & Wesselkamp (1997). Furthermore, the C_f and kurt values for the envelopes of these noises were also similar: C_f=3.84, kurt = 1.15 in Versfeld et al (2000); C_f=3.30, kurt = 1.05 in Kollmeier & Wesselkamp (1997); and C_f=3.48, kurt = 1.15 in the present study.

2. Although there are some standards aimed at optimizing the rms measurement of a speech signal (for example, ITU standard P.56 (1994) ‘Objective measurement of active speech level’) a classical method of rms calculation was employed in this study. The main reason for using the classical method of rms calculation was to compare the obtained results with the outcomes of the previous studies (Kollmeier & Wesselkamp, 1997; Versfeld et al, 2000). Furthermore, if the ITU norm had been used, it would be impossible to apply the Polish sentence material to standard clinical audiometers.

3. Equation (3) can also be used for a single sentence. According to this equation, a large S₅₀ of a sentence implies a comparable intelligibility of words constituting the sentence.

4. Our preliminary data show that, for the developed speech material, the approximate slope of psychometric functions for speech-shaped and CCITT maskers is 22%/dB and 14%/dB, respectively (Ozimek et al, 2006b). The following SRTs have been obtained: speech-shaped white noise −7 dB and CCITT −16 dB, respectively. Therefore, the spectral and temporal properties of the masking signal have a direct influence on S₅₀ and SRT values. Moreover, sensations produced by different maskers could influence the perception of speech under noisy conditions and, consequently, the parameters of the psychometric functions. An analysis of values of the perceptual parameters such as: loudness, sharpness, and roughness of the maskers used in different studies have shown that they are not equal. Definitions and significance of these parameters are given in Zwicker & Fastl (1999). Estimation of these parameters was performed using the ArtemiS v.4.00.300 software (HEAD acoustics GmbH) and was followed by normalization of the masker's level to 70 dB SPL. It has turned out that the masker used in the present investigation is characterized by higher loudness (G = 37.2 sone) in comparison with G = 31.8 sone, determined for the masker used by Versfeld et al (2000), and G = 33.9 sone, determined for the masker used by Kollmeier & Wesselkamp (1997). The so-called sharpness of maskers in the present study was S = 3.7 acum, while that in the study by Versfeld et al (2000) was S = 3.0 acum; and in the study by Kollmeier & Wesselkamp (1997) it was S = 2.7 acum. The roughness of the masker in the present study was R = 3.3 asper; while that in the paper by Versfeld et al (2000) was R = 2.8 asper; and in Kollmeier & Wesselkamp's study (1997), R = 3.2 asper.