基於音色與音高特徵之鳥鳴聲辨識方法研究

A Study of Bird Sound Recognition Based on Timbre and Pitch Features

廖偉恩(Wei-En Liao)

指導教授 ： 蔡偉和

國立臺北科技大學/電資學院/電腦與通訊研究所/碩士(2011年)

摘要

野外賞鳥已成為大眾休閒的新趨勢，但一般民眾常只能看見鳥或聽見鳥鳴聲，卻不知其種類為何。為了協助大眾識別鳥類，本論文探討鳥鳴聲自動辨識問題，透過音訊辨識相關技術，設計鳥鳴聲辨識系統。我們分別從音色及音高兩個層面進行分析，利用梅爾刻度倒頻譜係數表示鳥鳴聲的音色特徵，並搭配高斯混合模型進行特徵的參數模型化與比對；而音高層面分析則試圖求取鳥鳴聲所對應的音符，再利用雙連文模型捕捉音符的動態變化資訊，並據以比對未知鳥鳴聲。我
們挑選出大台北地區常見的十種鳥類，並從商業CD及鳥類相關網站上收集鳥鳴聲資料，使系統訓練和測試音檔分別屬不同的來源。實驗結果發現，採用音色、音高、與結合兩者的系統辨識正確率分別為 71.1%、72.1%、與75.04%。但若將鳥鳴聲分成鳴唱與鳴叫時，採用音色、音高、與兩階段式的系統辨識正確率分別為 95.02%、91.06%、與95.76%。

關鍵字

音色 ； 音高 ； 高斯混合模型 ； 雙連文模型

並列摘要

Wild bird watching has become a popular leisure activities in recent years. However, very often the public can only see birds or hear their sounds, but no idea about what kind of bird species they see. To help the public learn the bird species from their sounds, we propose an automatic system for recognizing birdsongs. Two acoustic features are used for analysis, timbre and pitch. In the timbre-based, Mel-scale cepstrum coefficients are used to characterizes the bird sound. Then, we use Gaussian Mixture Models represent the MFCCs as a set of parameters, so that it is easier to compare one bird species from another. In the pitch-based analysis, we convert bird sounds from their waveform representations into a sequence of MIDI note. Then, Bigram models are used to capture the dynamic change information of the notes.
We chose the top ten common bird species in the Taipei urban area. The audio data were collected from commercial CDs and the bird sound-related websites. We divided the data into non-overlaping two subsets, on for training and the other for testing. The results showed that the timbre-based, pitch-based, and the combination thereof system achieved 71.1%, 72.1%, and 75.04% accuracy of bird sound recognition, respectively. Further, if the bird sounds were pre-classified into calls and songs, the timbre-based, pitch-based, and the combination thereof systems achieved 95.02%, 91.06% and 95.76% accuracy of bird sound recognition, respectively.

並列關鍵字

Timbre ； Pitch ； Gaussian Mixture Model ； Bigram Model

參考文獻

[25] 楊青于，鳥聲辨識之初步研究與分析，碩士論文，國立清華大學，新竹，2005。

[1] 薛宇志，依照鳥類鳴叫與鳴唱聲識別其種類，碩士論文，國立台北科技大學，台北，2010。

[3] S. E. Anderson, A. S. Dave, and D. Margoliash, “Template-based automatic recognition of birdsong syllables from continuous recordings,” J. Acoust. Soc. Amer., vol. 100, no. 2, pp. 1209–1219, Aug. 1996.

[4] A. L. McIlraith and H. C. Card, “Birdsong recognition using backpropagation and multivariate statistics,” IEEE Trans. Signal Process., vol. 45, no. 11, pp. 2740–2748, Nov. 1997.

[5] J. Kogan and D. Margoliash, “Automated recognition of bird song elements from continuous recordings using dynamic time warping and hidden Markov models: A comparative study,” J. Acoust. Soc. Amer., vol. 103, no. 4, pp. 2187–2196, Apr. 1998.