本論文探討的內容主要關於人聲分類之研究。在第一個層次，我們將家庭環境中的聲音分為說話聲、人聲非說話聲以及環境音三種。為了達到此目標，我們尋找了許多特徵並找出合適的幾個作為參數。 在夜間睡眠研究的部份，我們希望可以將整晚的睡眠資料分為鼾聲與非鼾聲兩部份，針對鼾聲的部份再深入去探討是否有呼吸中止的情況產生。若受試者是在醫院或者專業睡眠實驗室收錄資料，則會有其他睡眠生理訊號可供分析。本論文試著找出鼾聲與振動、睡姿與睡眠階段的關係並有初步的成果。 / In this thesis, we describe the classification of audio signals in a smart home environment and in all-night sleep studies. In a home environment, our objective is different from most audio scene analysis projects in that we are mainly concerned with the distinction of human and non-human sounds. Toward this goal, we identify appropriate features to be extracted from the audio files and discuss the rationale behind choosing a particular feature. In all-night sleep recording, we describe the classification of audio signals recorded in all-night sleep studies. Our objective is to separate the episodes into snoring sounds and non-snoring sounds. We perform further analysis of the extracted snoring sounds to check if the testee has apnea. With polysomnogram data, we detect the relationship between snoring sounds and other sleep signals such as snoring vibration, sleep stages and body position.

人聲大致上可分為語音和非語音兩部分。傳統上對於聲音分類的研究大多強調語音、音樂和環境聲的分類。在本論文中，我們採取不同的觀點，著重於人聲中非語音部份的研究，聲音種類為笑聲、尖叫聲、打噴嚏聲和鼾聲。為了達到此目標，我們調查常用的幾種聲學特徵，並以多元適應性雲形迴歸和支持向量機進行特徵值篩選，找出對於非語音人聲分類具有代表性的聲學特徵。此外我們也進行多方面的模擬，以觀察雜訊對辨識率的影響。 本論文第二部份為鼾聲研究，我們比較一般普通麥克風和目前醫療用鼾聲麥克風(snoring microphone)、壓電感應器(piezo sensor)三者在偵測鼾聲上的表現。此外，並以KL divergence 和EMD兩種計算差異度的方法進行普通鼾聲與阻塞型鼾聲的分群。同樣地，我們加入不同程度雜訊至鼾聲訊號中，以測試兩方法抗雜訊的穩健度，結果顯示此兩種方法均有不錯的表現，其中EMD在大多數情況下有較佳的結果。 / Human sounds can be roughly divided into two categories: speech and non-speech. Traditional audio scene analysis research puts more emphasis on the classification of audio signals into human speech, music, and environmental sounds. We take a different perspective in this thesis. We are mainly interested in the analysis of non-speech human sounds, including laugh, scream, sneeze, and snore. Toward this goal, we investigate many commonly used acoustic features and select useful ones for classification using multivariate adaptive regression splines (MARS) and support vector machine (SVM). To evaluate the robustness of the selected features, we also perform extensive simulations to observe the effect of noise on the accuracy of the classification. / The second part of this thesis is concerned with the analysis snoring signals. We use ordinary microphone as our snoring recorder and compare its sensitivity with snoring microphone and piezo sensor, which are often utilized in clinical settings. In addition, we classify simple snores and obstructive snores using two distance measures: KL divergence and earth mover's distance (EMD). Similarly, we add noises to the snoring signals to examine the robustness of these two measures. It turns out that both methods perform satisfactorily, although EMD generates slightly better results in most cases.