基於時頻調變之適用於聽損病患的中文語音理解度客觀量測指標

Abstract

為了驗證助聽器演算法的效果，我們發展一套中文語音理解度量測指標來預估聽損者的語音理解度。為了符合不同聽損者的聽損程度，我們使用了一套能夠模擬聽損者耳蝸的聽損模型，模擬最小可聽水平提升、響度聚集、以及分頻解析度降低等聽損現象。許多文獻經由時域調變分析萃取出語音特徵來計算並預估語音理解度。在本論文中，我們採用時頻調變分析方法，擷取出影響語音理解度的語音特徵，來預估聽損者的語音理解度。時頻調變分析分為兩個聽覺感知階段，第一階段為人耳至中腦的頻譜預估，第二階段為中腦至大腦皮質聽覺區對時頻域調變的分析，同時考慮時間上與頻率上的變化。為了能考慮分頻解析度降低對語音理解度的影響，我們將參考Tiago H. Falk提出的基於時域調變封包為架構的理解度指標演算法，開發基於時頻調變的非侵入式語音理解度指標。設計實驗為量測聽損者在兩種雜訊環境下的中文單字語音理解度，並比較訊噪比對語音理解度的影響，最後比較Tiago H. Falk所提出的演算法及開發的理解度指標兩者與中文語音理解度的相關性，以評估兩種演算法的效果。未來將透過此開發的語音理解度客觀量測指標來開發語音增強演算法，並將其實際應用在助聽器。

In order to verify the performance of algorithms developed for hearing aids, we developed a measure that predicts the Mandarin speech intelligibility of the hearing impaired people. We constructed a model that simulates the cochlear of the hearing impaired to fit different patients. This model solves threshold elevation, loudness recruitment, and reduced frequency selectivity of hearing impaired. In order to predict speech intelligibility, many studies utilized the modulation spectral signal representation, which is obtained by an auditory-inspired filterbank analysis of the speech signal. In this study, a joint spectro-temporal auditory model was utilized to assess speech quality objectively. In this auditory model, the first stage is to simulate cochlear function of the spectrum estimation. The second stage is to simulate cortical function of the multi-dimensional spectrum analysis. In order to consider effecting speech intelligibility due to reduced frequency selectivity of hearing impaired, we developed a spectro-temporal modulations based non-intrusive intelligibility measure through referring to the construction of SRMR proposed by Tiago H. Falk. SRMR is a temporal modulation envelope based intelligibility measure. To validate our proposed measure, the performance of the proposed measure is compared to the SRMR intelligibility measurement algorithms under several noisy conditions. We can utilize the proposed measure to assess the performance of speech enhancement algorithms, and develop a speech enhancement algorithm in the future.