頻譜銳化對聽損者的感知影響研究

Abstract

實驗室先前所開發模擬聽損人士的個人化聽損模型，尚須經過驗證，我們與花蓮慈濟醫院合作，開發出可以測量聽損人士受損情況的系統，藉以得到病患的最小可聽水平、響度聚集以及聽覺濾波器的變寬程度。在測量過程中我們發現病患的聽覺濾波器變寬程度會隨著不同頻率而有不同的變化，因此，我們將學長的聽損模型做些改進後，讓正常聽力者聽到通過聽損模型的聲音與聽損人士所做的中文字認知辨識率達到相符合，成功的驗證了聽損模型的效用。接著在使用聽損模型之後我們發現，現有的助聽器雖對於消噪和語音的辨識可以達到不錯的效果，但是在於音樂上來說，這樣的處理過程反倒會讓音樂失去它原有的聲音。透過聽損模型我們可以聽到病患所聽的音樂會因為聽覺濾波器變寬而使得音樂變得模糊不清，故我們除了在不同頻率加上聲音強度的補償之外還利用了頻譜銳化來增強聽損者對於音樂的品質，並轉錄了多種曲風的音樂實際請病患測試並且評分，實驗結果確實達到有效的增進。

The previously developed hearing model for the hearing-impaired was not validated on actual patients. In this thesis, we collaborate with Hualien Tzu Chi Hospital to measure the broaden factor of each cochlear filter of hearing-impaired patients. The broaden factor combined with the hearing threshold and the degree of loudness recruitment establish the personalized hearing-impaired model. Mandarin intelligibility scores of each patient to normal speech are assessed and compared to the intelligibility scores of normal-hearing subjects to speech processed by the corresponding personalized hearing-impaired model. Comparable scores successfully validate our hearing-impaired model in predicting Mandarin intelligibility scores of patients. Algorithms in modern hearing aids are often developed to enhance speech perception, while inevitably deteriorating music perception by adding or subtracting frequency components. By examining the signals through the hearing-impaired model, we postulate that music heard by the hearing-impaired is blurred due to their broaden auditory filters. To compensate the broaden effect, we sharpen the spectrum of music to enhance music quality for the hearing impaired. Different styles of music clips are used to demonstrate that sharpening the spectrum does raise music quality scores of the hearing-impaired.