Frequency importance functions for words and sentences in Mandarin Chinese : implications for hearing aid prescriptions in tonal languages

Hearing aids increase the amount of speech perceived by the listener by amplifying the speech signal using a prescription that takes into account the type of hearing loss, and the frequency importance function (FIF) for the listeners’ language. An FIF describes how each frequency band contributes to speech intelligibility and is a reflection of the structure of the input language. Therefore, for Mandarin, the FIF may be affected not just by its phonemic structure, but also by the inclusion of low-frequency lexical tone. To date, most FIFs have been developed for English language background listeners. This project aims to fill a knowledge gap by providing a systematic evaluation of how the language features of Mandarin Chinese contribute to FIF for hearing aid prescription. Hence, this project uses series of speech recognition scores to develop Mandarin FIFs to examine weightings for lexical tone, phonetic content in words, and sentences, together with other functions to determine how the language factor (Mandarin Chinese) affects the estimation of speech intelligibility index (SII). Speech stimuli were recorded by a male broadcaster using monosyllabic words from Mandarin Speech Test Materials (MSTMs) and ten-word sentences from the Chinese Mandarin Hearing in Noise Test (CHINT). The speech was passed through a broadband digital filter to separate the speech signal into high- and low-pass conditions for discrete frequency bands (141, 224, 355, 562, 891, 1413, 2239, 3548, 5623, and 8913 Hz) and speech noise added to create 9 conditions for words and 7 conditions for sentences signal-to-noise ratios (SNRs: 15, 12, 9, 6, 3, 0, -3,-6,-9 dB). In the speech recognition test, normal hearing Mandarin speaking participants were asked to orally repeat words and sentences; and in the word session, to indicate the tone used on the screen. These data are used to determine performance intensity functions, the FIFs of tones, segments, words, and sentences. Lastly, the derived FIFs were incorporated with the published Long Term Average Speech Spectrum (LTASS) to evaluate SII transfer functions. Results of this project suggest that language specific inventories of phonemes and tones, in addition to context and redundancy effects, influence performance intensity functions, CFs, FIFs and TF slopes in Mandarin. The Mandarin CFs for the four speech types are remarkably similar for words (1807 Hz) and segments (1813 Hz), but lower for sentences (1570 Hz) and tones (743 Hz). FIF data showed a primary peak centered at ~1800 Hz and secondary peak at ~112 Hz for word and sentence materials. The Mandarin TFs indicate that tones, with high redundancy had the steepest slope, followed by sentences, then by words and segments, which both having similar shallow slopes. This study is the first tonal language study to separate the effects of words, segments and tones on performance intensity functions, FIFs, and TFs. The reported findings provide important new directions for developing hearing aid prescriptions for tonal language speakers, but specifically for Mandarin Chinese