Validation of the Articulation Index

Abstract

French and Steinberg proposed the basic concept and method of calculating the Articulation Index (AI) approximately 15 years ago. Although improvements and modifications of AI have been made over the years, it has not been generally accepted, perhaps because of insufficient evidence as to its validity. On the basis of studies reported in the literature and on new studies herein reported, it is shown that AI is a valid predictor of the intelligibility of speech under a wide variety of conditions of noise masking and speech distortion.

Spectral redundancy: Intelligibility of sentences heard through narrow spectral slits

“…The frequency importance functions derived from these measurements have been used for the construction of the Articulation Index, or AI (ANSI S3.5, 1969). The initial studies leading to the development of the AI were conducted at Bell Telephone Laboratories to provide a procedure that could be used to design and evaluate communication systems (see French & Steinberg, 1947;Kryter, 1962). The original frequency importance functions were developed for nonsense syllables, in part because ofthe sensitivity of functions based upon these stimuli to the effects of degrading speech.…”

Methods for assessing high frequency hearing loss in every-day listening situations

“…11, but there may be limitations. According to Stevens & Baruch (1957) and Kryter (1962), the Al-calculation may not always be valid for abruptly ending spectra. Nevertheless, it seems worthwhile to test the above mentioned hypothesis.…”

Extending the articulation index to account for non-linear distortions introduced by noise-suppression algorithms

“…The SNRs in each band are weighted by bandimportance functions which differ across speech materials (ANSI, 1997). The AI measure has been shown to predict successfully the effects of linear filtering and additive noise on speech intelligibility (e.g., Kryter, 1962aKryter, , 1962b.…”

“…The AI measure was further refined to produce the speech intelligibility index (SII) (ANSI 1997). The AI measure is based on the principle that the intelligibility of speech depends on the proportion of spectral information that is audible to the listener and is computed by dividing the spectrum into 20 bands (contributing equally to intelligibility) and estimating the weighted average of the signal-to-noise ratios (SNRs) in each band (Kryter, 1962a(Kryter, , 1962bPavlovic, 1987;ANSI, 1997;Amlani et al, 2002). The SNRs in each band are weighted by bandimportance functions which differ across speech materials (ANSI, 1997).…”