Abstract Voice recognition technology can almost accurately recognize a user’s voice in the absence of background noise or when the noise levels are extremely low. However, voice recognition in the presence of background noise with various voice signals is complicated. The problem in the current voice separation scheme is that though voice separation is possible from mixed voice signals, a permutation problem occurs that renders it difficult to identify the desired signal among the separated signals. In this paper, we propose a driver voice separation method for autonomous vehicles. To solve the permutation problem, active-beacon-based driver sound separation (ABDSS) utilizing active sound is used to distinguish the driver’s sound. After recording the voice work, simulation was performed. In the simulation, the proposed method succeeded in separating and distinguishing the original voice signals from the mixed voice signals. In addition, the coherence, kurtosis, and skewness calculation were used to verify that the separated signals were correctly identified in the simulation. Therefore, the proposed method is simpler in terms of hardware configuration than the existing methods and it is suitable for in-vehicle voice separation systems as well.

中文翻译：

---

用于自动驾驶汽车应用的基于主动信标的驾驶员声音分离系统

摘要 语音识别技术几乎可以在没有背景噪音或噪音水平极低的情况下准确识别用户的声音。然而，在各种语音信号存在背景噪声的情况下进行语音识别是复杂的。当前语音分离方案的问题在于，虽然可以从混合语音信号中分离语音，但是出现了置换问题，使得难以在分离的信号中识别所需信号。在本文中，我们提出了一种用于自动驾驶汽车的驾驶员语音分离方法。为了解决置换问题，基于主动信标的驾驶员声音分离（ABDSS）利用主动声音来区分驾驶员的声音。录制完语音工作后，进行了模拟。在模拟中，所提出的方法成功地从混合语音信号中分离和区分了原始语音信号。此外，还使用相干性、峰态和偏度计算来验证在模拟中正确识别了分离的信号。因此，所提出的方法在硬件配置方面比现有方法更简单，并且也适用于车载语音分离系统。