透過您的圖書館登入
IP:44.204.217.37
  • 學位論文

雙耳量測聽音辨位技術研究

The binaural measurements for identification of source position

指導教授 : 洪振發

摘要


本文研究空間聽覺定位,其中包含雙耳線索(Binaural cues)及頻譜線索(Spectral Cues)。雙耳線索分為雙耳時間差(Interaural Time Difference, ITD)與雙耳聲強差(Interaural Level Difference, ILD),頻譜線索則為人類身體構造對聲壓頻譜的增益特性,類似於濾波器,可用頭部關係轉移函數(Head Related Transfer Function, HRTF)表示。雙耳線索有其頻寬的侷限性及部分方位模糊等問題。頻譜線索反應頭部及軀幹等對聲音訊號的遮蔽作用,對於不同方位的聲源其頻譜也有不同變化,其中包含的方位訊息可能是人耳能分辨方位的重要因素。 本文研究麥克風陣列聲源定向的方法,包含多重訊號分類演算法(Multiple Signal Classification, MUSIC)、麥克風陣列語者定位演算法(Speaker Localization by a Arrayed Microphone, SLAM)兩種演算方法。透過電腦模擬聲源訊號並實際架設麥克風陣列,使用NI-DAQ將聲音訊號擷取置電腦,並使用NI LabVIEW及MATLAB等軟體進行MUSIC及SLAM演算法,估算出聲音訊號方向。實驗結果顯示在安靜的室內能對方位角做有效的識別。 不過麥克風陣列仍無法解釋人只須兩耳即可辨識聲源方向。因此接著研究頻譜線索的聲源定向方法,使用頭部關係轉移函數做為主要識別依據,建立類神經網路映射模型,並使用MIT HRTF資料庫,透過不同的方位編碼方式進行聲源方位的辨識,實驗結果顯示對於方位角及俯仰角皆有不錯的識別結果。

並列摘要


In this thesis,we study spatial hearing which including binaural cues and spectral cues for sound localization. Binaural cues have some drawbacks: ITD and IID only work on their suitable frequenccy range and have problems on the ‘cone of confusion’.The spectal cues, HRTF ( Head Related Transfer Function) points out the acoustic filtering effects on receving sound from the sound source to the head and torso. Sound source from different spatial angles have different spectrals too. Two sound localization methods by microphone arrray are investigated, MUSIC ( Multiple Signal Classification) algorithm and SLAM ( Speaker Localization by a Arrayed Microphone). We use computer to simulate sound signals and then construct particle microphone arrays, which use a DAQ ( Data acquisition) to transmit sound signal to a computer, and use NI LabVIEW and MATLAB to operate MUSIC and SLAM. The experiment indicate the sound localization results are reliable in the quiet room. In addition, we study spectral cues for sound localization because of microphone array cannot expalin why human can determine the direction of sound source by only two ears. Using HRTF as the main recognization cue, then building neural network model with MIT HRTF database, different coded directional informations are headed to do sound localizations. The results indicate good accuracy on azimuth and elevation angle recognization.

並列關鍵字

sound localization microphone arrray HRTFs

參考文獻


[16] Sacha Barber, 2007, AI : Neural Network for beginners.
[1] Blauert, J., 1982. Spatial Hearing, The Psychophysics of Human Sound Localization.The MIT Press, Cambrige, Massachusetts.
[2] Stevens, S. S., and Newman, E. B., 1936. The Localization of Actual Sources of
Sound, America Journal of Psychology.
[3] Ephraim, Y., and Van Trees, H. L., 1995, A signal subspace approach for speech

延伸閱讀