聲音辨識的研究在學術界與業界均行之有年,許多研究人員努力的辨識出更複雜的聲音。聲音辨識往往是利用關聯或規則,尋找出聲音資訊中所含的意義。電腦有了聲音辨識的技術之後,便可以由人的聲音當作指令再經過電腦的判斷,來完成人類所交待的事項。只不過這些聲音的研究大多在於人類的語音或音樂,對於人們生活周遭中必定會遇到的,救護車鳴笛聲辨識卻是屈屈可數。 本論文提出一種辨識救護車聲音的方法,該方法以Support Vector Machine做為分類的基礎,分為訓練過程與辨識過程。訓練過程中,由乾淨的救護車鳴笛聲萃取進行訓練,並以救護車鳴笛聲低頻600-750Hz、高頻900-1050Hz為聲音的特徵,結合Support Vector Machine作為聲音辨識方法。在辨識過程中,使用在道路上錄製的聲音檔,分別有汽、機車及救護車混合的聲音和單純汽、機車的聲音。將聲音檔以0.1秒切割成一個frame,使用快速傅利葉轉換將時域資料轉為頻域資料,截取聲音中我們所要的頻率特徵,利用訓練所得到的分類模型,來判斷是否為救護車鳴笛聲。
Study the implementation of voice recognition for a long time, many researchers to identify more complex sound. Use of voice recognition is usually associated with, or rules, to find out the meaning of the sound. Computer voice recognition techniques, the computer can use voice control. However, almost all of these voices voice or music, for the road will have to face sound of ambulance siren is very little recognition. This paper presents an ambulance voice recognition method to Support Vector Machine as the basis of classification, the process is divided into the training process and the identification process. Training process, from the clean sound of an ambulance train whistle. Ambulance siren sound low frequency 600-750Hz, high-frequency 900-1050Hz for voice features, combined with Support Vector Machine as a voice recognition method. In the identification process, the use of recorded sound on the road. Cars ambulance siren sound mixing, and pure cars sound. 0.1 seconds the sound file to cut a frame, using the FFT to time domain data into frequency domain data. The frequency of interception we want to use the training received classify model, identify whether there is an ambulance siren sound.