應用最長共同子序列與模糊方法辨識救護車鳴笛聲

Using Longest Common Sequence and Fuzzy to Recognize the Ambulance Siren Sound

王文聖(Wen-sheng Wang)

指導教授 ： 廖俊鑑

朝陽科技大學/資訊學院/資訊與通訊系碩士班/碩士(2013年)

摘要

聲音辨識是利用聲音訊號的特徵，獲得聲音資訊中所包含的意義。多數聲音的研究在於人類的語音或音樂的節奏，對於攸關生命安全的救護車鳴笛聲辨識卻是沒有太多著墨。救護車鳴笛聲的功能在於提醒其他用路駕駛即時做出讓道的回應。但實際情況下，救護車鳴笛聲會受噪音破壞及聲音阻隔，使駕駛無法聽到救護車聲音。本論文提出一個救護車鳴笛聲辨識方法，可用於偵測到救護車聲音時，判斷聲音訊號是否為救護車鳴笛聲，進而適時提醒駕駛者有救護車接近。

救護車鳴笛聲的頻率與持續時間於消防署皆有明確的規範，規範定義高頻頻帶900Hz至1000Hz，低頻頻帶650Hz至750Hz，高頻持續時間0.6秒，低頻持續時間0.4秒，高、低頻兩者規律交互組成。本研究便根據這項頻率特性，使用最長共同子序列(Longest Common Subsequence, LCS)演算方法，並藉由運算後的LCS輸出數據，進一步使用模糊判斷方法，判別聲音訊號中是否有救護車鳴笛聲存在。並由我們實驗結果顯示，在不同噪音環境下，系統於偵測到救護車鳴笛聲時，平均正確率為92.8%，且無偵測到救護車鳴笛聲時，平均正確率為99.8%。

關鍵字

救護車鳴笛聲 ； 模糊方法 ； 最長共同子序列

並列摘要

Sound recognition utilizes sound signal characteristics to obtain meanings encompassed in the sound information. Study of sound focuses mostly on the tempo of human language or music but little on the recognition of ambulance siren which is critical to the safety of life. The function of an ambulance siren is to remind other drivers on the road to make timely response to give way to the ambulance. However, the siren of an ambulance is often disrupted by noise and blocked by voices in a real life situation. This prohibits other drivers from hearing the siren. This research presents a method to recognition ambulance siren. Upon the detection of an ambulance sound, it can be utilized to verify if the sound signal is from an ambulance siren and then to remind drivers in a timely way that an ambulance is approaching.

The National Fire Agency has specific regulation on the frequency and duration on ambulance siren. This regulation defines high frequency of between 900 Hz to 1000Hz and low frequency of between 650Hz and 750Hz with 0.6 second duration for high frequency and 0.4 second duration for low frequency. The sound of siren is composed of regular alternating sound between high and low frequencies. Based on this frequency characteristic, this research utilizes the Longest Common Subsequence (LCS) algorithm. With LCS output statistics, it further uses fuzzy judgment method to verify if ambulance siren exists in the sound signal. Our experiment result indicates that, under environments with different noises, the average true positive rate is 92.8% when the system detects ambulance sound and 99.8% when the system doesn''t detect ambulance sound.

並列關鍵字

Ambulance Siren Sound ； Fuzzy ； Longest Common Sequence

參考文獻

[7] 黃全淯，車載感測網路於大範圍環境監測二氧化碳之應用，碩士論文，國立交通大學資訊科技產學專班，新竹，2008。

[1] P. Fan, “Improving Broadcasting Performance by Clustering with Stability for Inter-Vehicle Communication,”in Proc. of 65th Vehicular Technology Conference, pp. 2491-2495, 2007.

[4] T. H. Ho and S. J. Chung, “A compact 24 GHz radar sensor for vehicle sideway-looking applications,” Microwave Conference, European, pp. 351-354, 2005.

[10] B. Milner, and X. Shao, “Clean speech reconstruction from MFCC vectors and fundamental frequency using an integrated front-end,” Speech Communication, Volume. 48, Issue 6, pp. 697-715, 2006.

[11] S. Imai, “Cepstral analysis synthesis on the mel frequency scale,” Proc. of ICASSP, volume. 8, pp. 93–96, 1983.