Development of a Danish speech intelligibility test

Abstract

A Danish speech intelligibility test for assessing the speech recognition threshold in noise (SRT_N) has been developed. The test consists of 180 sentences distributed in 18 phonetically balanced lists. The sentences are based on an open word-set and represent everyday language. The sentences were equalized with respect to intelligibility to ensure uniform SRT_N assessments with all lists. In contrast to several previously developed tests such as the hearing in noise test (HINT) where the equalization is based on scored (objective) measures of word intelligibility, the present test used an equalization method based on subjective assessments of the sentences. The new equalization method is shown to create lists with less variance between the SRT_Ns than the traditional method. The number of sentence levels included in the SRT_N calculation was also evaluated and differs from previous tests. The test was verified with 14 normal-hearing listeners; the overall SRT_N lies at a signal-to-noise ratio of −3.15 dB with a standard deviation of 1.0 dB. The list-SRT_Ns deviate less than 0.5 dB from the overall mean.

• Speech intelligibility assessment
• Equalization of intelligibility
• HINT
• Speech perception

Abbreviations
CLUE	=	Conversational language understanding evaluation
HINT	=	Hearing in noise test
HL	=	Hearing level
RMS	=	Root-mean-square
SNR	=	Signal-to-noise ratio
SRTN	=	Speech recognition threshold in noise
SPL	=	Sound pressure level
SI	=	Sentence intelligibility
WI	=	Word intelligibility

Abbreviations
CLUE	=	Conversational language understanding evaluation
HINT	=	Hearing in noise test
HL	=	Hearing level
RMS	=	Root-mean-square
SNR	=	Signal-to-noise ratio
SRTN	=	Speech recognition threshold in noise
SPL	=	Sound pressure level
SI	=	Sentence intelligibility
WI	=	Word intelligibility

Sumario

Se ha desarrollado una prueba danesa de inteligibilidad del lenguaje para evaluar el umbral de reconocimiento del lenguaje en ruido (SRT_N). La prueba consta de 180 frases distribuidas en 18 listas fonéticamente balanceadas. Las frases están basadas en palabras de contexto abierto y representan el lenguaje cotidiano. Las frases fueron ecualizadas con respecto a la inteligibilidad para asegurar una evaluación uniforme del SRT_N en todas las listas. En contraste con varias pruebas previamente desarrolladas, tales como la prueba de audición en ruido (HINT), donde la ecualización está basada en medidas calificadas (objetivas) de inteligibilidad de la palabra, esta prueba usó un método de ecualización basado en evaluaciones subjetivas de las frases. El nuevo método de ecualización parece crear listas con menos variancia entre el SRT_NS que el método tradicional. El número de niveles de frase incluidos en el cálculo del SRT_N también fue evaluado y difiere de pruebas previas. La prueba fue verificada con 14 sujetos normo-oyentes; el SRT_N global se encuentra con una relación señal-ruido de -3.15 dB con una desviación estándar de 1.0 dB. El SRT_NS de la lista se desvía menos de 0.5 dB de la media global.