时频特征的海豚发声端点检测方法研究*

为在复杂的海洋环境噪声场中检测出目标海豚的某类声信号，设计了一种基于海豚声学特征的端点检测方法。根据海洋环境中噪声能量大且分布频率范围广，而海豚声音的时频特征具有持续时间短，、频率高且集中，、发声行为持续时间长的特点，将采集到的海洋声音进行分帧，计算单帧信号短时能量、谱质心和谱质心二阶偏移率，当海豚发出声信号时，谱质心和能量相应发生突变，截取发生突变的信号实现端点检测。通过与门限法等其他常见端点检测方法进行对比，结果表明，此方法在低信噪比的海洋环境下对单一海豚的某一类声信号检测准确度更高，同时具有较强的抗干扰能力。

The detection method of dolphin vocal endpoint based on time-frequency characteristics

In order to detect a certain kind of sound signal of target dolphins in the complex marine environment noise field, an endpoint detection method based on dolphin acoustic characteristics was designed. According to the large noise energy and wide distribution frequency range in the marine environment, the time-frequency characteristics of dolphin sound have the characteristics of short duration, high frequency and concentration, and long duration of phonation behavior. The collected ocean sound is divided into frames, and the short-time energy, spectral centroid and second-order shift rate of the single frame signal are calculated. When the dolphin sends out the sound signal, the spectral centroid and energy change correspondingly, and the signal with the change is intercepted to realize endpoint detection. Compared with other common endpoint detection methods such as threshold method, the results show that this method has higher accuracy in detecting a certain kind of sound signal of a single dolphin in the marine environment with low signal-to-noise ratio, and has strong anti-interference ability.