本研究利用現有語音信號處理常用的幾個方法,如快速傅利葉轉換、線性預估、以及在頻譜上的高低頻能量分析,作為此一研究題目的幾個重要的方法。主要是以這些在時域與頻域上的分析方法,將"語音訊號處理"與"中醫聞診"兩者結合,並整合到Apple公司的iPad裝置上,透過iPad的錄音設備,將語音紀錄下來分析。本研究比較所有受事者的資料,發現正常人的聲音比氣虛患者的聲音更為穩定,在長時間發音10秒時特別明顯,從正常人第一秒到第八秒由LPC頻譜看起來,共振峰的位置幾乎沒什麼改變,第一與第二和第三共振峰位置明顯,語音的強度波形也幾乎沒什麼太大的變化。而氣虛患者的發音,因為無法維持語音波形的平穩,會出現較大的波形改變,往往第二共振峰會不是那麼明顯,甚至有可能會消失。另一方面,在區域能量的比較上,正常人較虛症患者來的集中且穩定。藉由這些不同設計出2個不同的特徵,用”費雪分類”這個方法來區分出正常人與虛症患者的不同,結果顯示正常人的辨識率能達到92%,氣虛患者的辨識率能達到94.4%,並期望結果能為患者提供日常居家生活自我檢測的參考,或是成為中醫醫生在看診時的參考器材,為醫界盡一份心力。 關鍵字:中醫聞診、氣虛、iPad、語音訊號處理
This study used the methods of digital signal processing such as fast Fourier transformation, linear prediction coding, short-time energy, and power spectrum to analyze the recorded data in this study. The combination of Chinese medicine and automatic speech recognition technology based on major analysis methods of time-domain, frequency-domain and integrated interface with Apple iPad system. The study recorded voice for analysis by using iPad. The study compared all of the data and found that the voice signals of healthy subjects showed to be more stable than unhealthy subjects. This characteristic was significant when testing the voice lasting recorded for 10 seconds. Healthy subjects did not show difference from the start to the end of the recorded voice. However, unhealthy subjects had difficulty of maintaining a stable voice. Their voices presented various formants around the end of lasting voice. Since unhealthy subjects could not maintain stable voice, their LPC spectrums drifted or disappeared. In addition, healthy subjects were more stable than unhealthy subjects in the special region of power spectrum. This study designed two features and used the fisher discriminate methods to analyze and discriminate subject’s health. The result showed the mean accuracy of 92% in health subjects and 94.4% in unhealthy subjects. The study could be applied as a daily self-monitoring tool to provide useful information about health conditions or as a reference of diagnosis. Keywords: Qi vacuity、iPad、speech signal process