The band importance function for the Dutch digits-in-noise test: A computational derivation using human and automatic speech recognition data, including extended high frequencies.

Abstract

The band importance function (BIF) quantifies the contribution of each frequency band to the intelligibility of speech and is a key element of the speech intelligibility index. The BIF depends on the type of speech material and speaker characteristics. Several BIFs exist, but they do not cover all test materials used in audiology. This study aimed to determine the BIF for the Dutch digits-in-noise (DIN) test, a widely used speech-in-noise measure. Twenty young adults with normal hearing completed a speech recognition task using Dutch DIN triplets presented at multiple signal-to-noise ratios (SNRs) and filtered according to one-third octave bands. The BIF was estimated using the computational method of Kates (2013), applying both digit and triplet scoring approaches. In addition, two automatic speech recognition (ASR) systems, Whisper and FADE, were used to generate comparative BIF estimates. Extended high frequencies (EHFs) were included to assess their contribution to intelligibility. The results showed that the BIF of the Dutch DIN test substantially deviates from other BIFs, particularly in the contribution of high frequencies. Approximately 9% of speech information was attributed to the EHFs. Validation showed improved prediction of DIN test speech recognition thresholds using the newly derived BIF compared to the SPIN BIF. Scoring method influenced band importance slightly, whereas ASR-based estimates did not match human-derived BIFs.