Using the Speech Transmission Index to predict the intelligibility of non-native speech
Sander J. van Wijngaarden, Herman J. M. Steeneken, Tammo Houtgast, Adelbert W. Bronkhorst · The Journal of the Acoustical Society of America · 2002
The calibration of the Speech Transmission Index (STI) is based on native speech, presented to native listeners. This means that the STI predicts speech intelligibility under the implicit assumption of fully native communication. In order to assess effects of both non-native production and non-native perception of speech, the intelligibility of short sentences was measured in various non-native scenarios, as a function of speech-to-noise ratio. Since each speech-to-noise ratio is associated with a unique STI value, this establishes the relation between sentence intelligibility and STI. The difference between native and non-native intelligibility as a function of STI was used to calculate a correction function for the STI for each separate non-native scenario. This correction function was applied to the STI ranges corresponding to certain intelligibility categories (bad–excellent). Depending on the proficiency of non-native talkers and listeners, the category boundaries were found to differ from the standard (native) boundaries by STI values up to 0.30 (on the standard 0–1 scale). The corrections needed for non-native listeners are greater than for non-native talkers with a similar level of proficiency. For some categories of non-native communicators, the qualification excellent requires an STI higher than 1.00, and therefore cannot be reached.