Intelligibility of speech is important for comprehending the amount of speech information that is both available and audible to a receiver during disruptive conditions that could prevent comprehension of a spoken message.
Speech intelligibility index (SII) is a measure, between 0 and 1, that represents the intelligibility of speech under a variety of adverse listening conditions, such as noise masking, filtering, and reverberation. Speech cues interrupted by fewer of these conditions will be more available to the listener, and will thus have a higher SII value.
First established by ANSI/ASA S3.5-1997 (R2020) – American National Standard Methods for Calculation of the Speech Intelligibility Index, SII is defined as the “product of band importance function and band audibility function, summed over the total number of frequency bands”. In symbols, this can be understood as:
Where n is the number of SII computational bands, while Iiand Ai are the values of the band importance function and the band audibility function associated with the frequency band designated by the summation index i. The band referred to is a frequency band, which designates the high and low frequencies in which a sound is emitted. This concept is essential for establishing the basis of speech in its regular conditions.
Workers at Bell Telephone Laboratories in the 1960s conducted the earliest examination of the interaction of different noises influencing speech-recognition performance. Their extensive work led to the definition of the acoustical index, known as the articulation index (AI). The calculation for the AI was first presented in the original version of the S3.5 standard, ANSI/ASA S3.5-1969.
However, the 1997 revision of the document replaced articulation index with speech intelligibility index. While these two terms are very similar, actually deriving from the same philosophy of understanding disruptions and their influence on the delivery of speech, they contain some differences that made SII more suitable for the newer version of the document. Specifically, the standard now allows for the input of various variables, due to the support of its wider framework. This makes SII more applicable because AI is incapable of handling the necessary variety of inputs.
ANSI/ASA S3.5-1997 (R2020) – American National Standard Methods for Calculation of theSpeech Intelligibility Index addresses the method for calculating the physical measure of speech intelligibility index. The document provides the means by which a user can estimate all of the input values, such as equivalent speech spectrum level, equivalent noise spectrum level, and equivalent hearing threshold level, that become part of the final calculation.
Through the guidance of ANSI/ASA S3.5-1997 (R2020), SII may be computed through four different methods: critical frequency band, one-third octave frequency band, equally contributing critical band, and octave frequency band. The requirements and formulas needed for carrying out these calculations are addressed in the standard.
The applications of this standard are plentiful, as it is stated in the document that it extends to all listening conditions where the specified input variables exist. This has many different uses, such as in research, e.g. for determining the impact that hearing loss has on the audibility of speech, or for testing PA systems, which can be present in a variety of venues, public or private.
Due to the complexity of the topic, there are three different programs available for calculating speech intelligibility index that make use of the guidelines covered in this standard, However, before using the software, the Acoustical Society of America (ASA) recommends that users familiarize themselves with the standard. The programs can be downloaded from here: Programs for SII
1. Acoustical Society of America (ASA), ANSI/ASA S3.5-1997 (R2017) – American National Standard Methods for Calculation of the Speech Intelligibility Index (New York: ASA, 2017), 11.
