基于计算听觉场景分析(Computational Auditory Scene Analysis,CASA)的语音分离系统通过模拟人耳的听觉感知系统对混合信号进行处理并分离出感兴趣的目标语音,近年来得到了很大的发展。如何在干扰噪声存在的情况下进行正确的基音提取...基于计算听觉场景分析(Computational Auditory Scene Analysis,CASA)的语音分离系统通过模拟人耳的听觉感知系统对混合信号进行处理并分离出感兴趣的目标语音,近年来得到了很大的发展。如何在干扰噪声存在的情况下进行正确的基音提取跟踪一直是CASA系统研究的重点。提出了一种基于目标语音源的改进基音跟踪算法。该算法通过对目标源估计和基音检测两个步骤的反复迭代计算,得到最终的基音轨迹。通过在不同噪声干扰条件下与传统基音跟踪算法对比的实验结果证明,该算法能够有效地抑制噪声,提高输出语音的信噪比和语音质量。展开更多
The double pulse sources （DPS） method is presented for linear track estimation in this work. In the field of noise identification of underwater moving target, the Doppler will distort the frequency and amplitude of ...The double pulse sources （DPS） method is presented for linear track estimation in this work. In the field of noise identification of underwater moving target, the Doppler will distort the frequency and amplitude of the radiated noise. To eliminate this, the track estimation is necessary. In the DPS method, we first estimate bearings of two sinusoidal pulse sources installed in the moving target through baseline positioning method. Meanwhile, the emitted and recorded time of each pulse are also acquired. Then the linear track parameters will be achieved based on the geometry pattern with the help of double sources spacing. The simulated results confirm that the DPS improves the performance of the previous double source spacing method. The simulated experiments were carried out using a moving battery car to further evaluate its performance. When the target is 40-60m away, the experiment results show that biases of track azimuth and abeam distance of DPS are under 0.6° and 3.4m, respectively. And the average deviation of estimated velocity is around 0.25m/s.展开更多
文摘基于计算听觉场景分析(Computational Auditory Scene Analysis,CASA)的语音分离系统通过模拟人耳的听觉感知系统对混合信号进行处理并分离出感兴趣的目标语音,近年来得到了很大的发展。如何在干扰噪声存在的情况下进行正确的基音提取跟踪一直是CASA系统研究的重点。提出了一种基于目标语音源的改进基音跟踪算法。该算法通过对目标源估计和基音检测两个步骤的反复迭代计算,得到最终的基音轨迹。通过在不同噪声干扰条件下与传统基音跟踪算法对比的实验结果证明,该算法能够有效地抑制噪声,提高输出语音的信噪比和语音质量。
文摘The double pulse sources （DPS） method is presented for linear track estimation in this work. In the field of noise identification of underwater moving target, the Doppler will distort the frequency and amplitude of the radiated noise. To eliminate this, the track estimation is necessary. In the DPS method, we first estimate bearings of two sinusoidal pulse sources installed in the moving target through baseline positioning method. Meanwhile, the emitted and recorded time of each pulse are also acquired. Then the linear track parameters will be achieved based on the geometry pattern with the help of double sources spacing. The simulated results confirm that the DPS improves the performance of the previous double source spacing method. The simulated experiments were carried out using a moving battery car to further evaluate its performance. When the target is 40-60m away, the experiment results show that biases of track azimuth and abeam distance of DPS are under 0.6° and 3.4m, respectively. And the average deviation of estimated velocity is around 0.25m/s.
