An objective identification of spectral distinctiveness on acoustic cue to subjects with hearing loss

Hearing loss would seriously degrade a subject's speech perception, thereby affecting the development of articulation ability. Assistive listening devices such as hearing aids or cochlear implants could help subjects with hearing loss develop their speech perception. To facilitate this process, speech-language patholo-gists (SLPs) provide speech-perception training that can increase the subject's ability to distinguish one phoneme from another. In clinical practice, they manually enhance the acoustic cues and use this information therein to teach a subject the differences between two phonemes. However, it is a time-consuming and expensive process. In this study, an objective identification of spectral distinctiveness is proposed to find the acoustic cues between two phonemes. SLPs then could use it to facilitate the speech-perception training process. Mel-frequency cep-strum coefficients (MFCCs) are selected to accurately represent the characteristics of acoustic signal. To deal the mismatch between two phonemes in time domain, the Viterbi algorithm is integrated to find a best matching condition. In order to estimate the spectral similarity, the speech signal is decomposed into different frequency bands. For each bands, MFCCs are also applied to represent the characteristic of filtered speech signals. According to the best matching condition, the spectral distinctiveness of different bands between two phonemes could be easily estimated by using Euclidean metrics. Finally, the morphological gradient is adopted to find the sections of spectral distinctiveness. To evaluate the correction of proposed approach, the speech signal of identified spectral distinctiveness was manually manipulated, and then the manipulated speeches are evaluated by a close-set detection evaluation methodology. The experimental results demonstrated that our approach is able to identify a suitable spectral distinctiveness.