Methods for the Calculation and Use of the Articulation Index

Abstract

Speech-intelligibility testing is an expensive and time-consuming operation that requires laboratory test conditions. In an attempt to short-cut or make unnecessary this type of testing, a procedure was developed by French and Steinberg [J. Acoust. Soc. Am. 19, 90–119 (1949)] for calculating from physical and acoustical measurements made on a communication system a measure that is indicative of the intelligibility scores that would be obtained for that system under actual test conditions. This measure is called the “Articulation Index” (AI). Methods of calculating AI have been improved and elaborated to the point where several methods for its calculation, herein reported, can be proposed for use in the evaluation of most speech communication systems.

A review of the perceptual effects of hearing loss for frequencies above 3 kHz

“…For some specific speech tests, using nonsense syllables where each English phoneme occurs equally often, CID words, NU6 nonsense syllables, the diagnostic rhyme test, short passages of easy materials, and SPIN test monosyllables, the corresponding percentages are 26, 16, 17, 17, 18, and 20%, respectively. When the face of the talker is visible, so lip-reading is possible, the high-frequency components in the acoustic signal become relatively less important (Kryter, 1962;Sumby & Pollack, 1954). However, there are many situations when lip-reading is not possible, for example, when listening to a companion at dinner while cutting up food or when listening to the radio.…”

“…In these studies, speech has been highpass or lowpass filtered with various cutoff frequencies, and speech intelligibility has been measured for each cutoff frequency. Such studies formed the basis for the Articulation Index (ANSI, 1969;Fletcher, 1953;French & Steinberg, 1947;Kryter, 1962) and its successor, the SII (ANSI, 1997) that is described in the next section of this paper. For example, French and Steinberg showed that decreasing the cutoff frequency of a lowpass filter from 7 to 2.85 kHz decreased the percentage of correctly identified syllables presented in quiet from 98 to 82%.…”

A data-driven non-intrusive measure of speech quality and intelligibility

“…Objective intelligibility assessment methods operate without the need for human subjects. One of the earliest intrusive intelligibility metrics was proposed by French and Steinberg [33] as the Articulation Index (AI), which was further refined with improved methods for calculating the AI by Kryter in 1962 [34] and finally led to an ANSI standard in 1969 [35]. The AI was further developed into the Speech Intelligibility Index (SII) and led to an ANSI standard for intrusive intelligibility assessment in 1997 [36].…”

An efficient coding of the prediction residual

“…However, the speech intelligibility aspects, such as Articulation Index (Al) [29], have not been used in these. The Articulation Index concept has been used effectively in the sub-band coding of speech [36].…”

“…Noting the speech characteristics in the residual signal, several authors have investigated the coding aspects of the prediction residual [13][14][15][16][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31][32]. However, the speech intelligibility aspects, such as Articulation Index (Al) [29], have not been used in these.…”