中文说明：在声学事件检测中， 采用基于短时能量和短时过零率的端点检测方法和基于动态时间规整算法的情况下， 对鸟叫声、 人说话声和车辆驶过声这三类声学事件进行检测。 在不加噪声的情况下， 基于时域特征的识别准确率是88.89%， 基于频域的识别准确率是83.33%， 基于时频域特征的识别准确率是77.78%。 由上述分析可知， 在不加噪声的情况下， 基于时域特征检测的识别精确度是最好的。 在加噪声的情况下， 分别基于时域特征和频域特征的检测， 噪声对声学事件的干扰与覆盖影响较为重大， 噪声强度越增加， 对系统的检测性能所受到的影响也随之俱增。 对于基于时频域特征的检测而言， 噪声的影响并不十分明显， 且耗时少， 具有很好的应用前景。 所以， 在加噪声的情况下， 基于时频域特征检测的识别精确度是最好的。
English Description：
In acoustic event detection, the endpoint detection method based on short-term energy and Short-term zero crossing rate and the dynamic time warping algorithm are used to detect three kinds of acoustic events: bird calls, human voices and vehicle passing sounds. Without noise, the recognition accuracy based on time domain feature is 88.89%, the recognition accuracy based on frequency domain feature is 83.33%, and the recognition accuracy based on time-frequency domain feature is 77.78%. From the above analysis, it can be seen that the recognition accuracy based on time domain feature detection is the best without noise. In the case of adding noise, the detection based on time-domain characteristics and frequency-domain characteristics respectively, the noise has a significant impact on the interference and coverage of acoustic events. The more the noise intensity increases, the more the impact on the detection performance of the system increases. For the detection based on time-frequency characteristics, the influence of noise is not very obvious, and it takes less time, so it has a good application prospect. Therefore, in the case of adding noise, the recognition accuracy based on time-frequency feature detection is the best.