聲音辨識技術一直是一個很重要的課題，因為其發展使我們的生活更加便捷，並且，近年來此項技術也被廣泛應用在一些移動裝置如:智慧型手機、平板等等。因此，如何開發一套效果良好之音訊辨識系統非常重要。聲音可細分為很多種類型，在這篇論文中，我們針對環境聲音事件來研究。 我們提出的辨識系統以傅立葉轉換為基礎，結合了所提出之動態Local Binary Pattern (LBP) Uniform與具平滑化功能之Filter，並且利用Variance Measure (VAR)作為前處理來強化時頻圖之邊緣紋理與對比度。 在我們提出的系統中，利用Box Filter 與Gaussian Filter來使傅立葉轉換後的時頻圖平滑。此外，我們進一步考慮到時頻圖中能量分布差異的特性，提出了動態Local Binary Pattern (LBP) Uniform方法。本論文提出把頻譜圖分為不同頻段區域，並且藉由對LBP Histogram降維來動態的調整不同頻率之解析度，以形成特徵參數並藉由Support Vector Machine(SVM)來進行環境聲音辨認。;Sound recognition has become an important application in some devices. The type of sound to be recognized may vary, e.g., musical instrument sounds, environmental sounds, and speech. In this study we use environmental sound for our experiment. Time-frequency, which can represent an audio signal, is a form of texture image that can be used for image classification. In this paper, we introduce a simple image classification method using local binary pattern (LBP) and an image smoothing method prior to feature extraction to reduce spectrogram image noise. In this thesis, we combine spectrograms and LBP uniform with an image filter and variance measure (VAR) for contrast enhancement. We alsointroduce adynamic LBP method to reduce the dimension in difference dimension for each sub-band(high, middle, and low frequency). After using image filter as pre-treatment and VAR for contrast enhancement, weconcatenate all thesefeatures. To remove image noise, we use two types of smoothing filter:a box filter (mean filter) and a Gauss filter. To improve recognition, filtering is applied as a pretreatment prior to feature extraction. To enhance local image texture contrast, such as object edges and corners, we use a VAR function. We use a support vector machine for the classifier.
