Scholarlink

<p>Sound source localization (SSL) plays a fundamental role in spatial audio processing and human–machine interaction. Most existing SSL studies focus on estimating sound source directions in a two-dimensional plane, typically limited to azimuth-only localization. However, in many practical applications such as smart home devices, voice-controlled assistants, and indoor service robots, three-dimensional (3D) directional localization is required. Therefore, this study proposes a sound-intensity-based approach for estimating three-dimensional sound source directions, characterized by azimuth and elevation angles, using a differential microphone array (DMA). Specifically, a Tri-axis Sound Intensity Spectrogram (TriSIS) is first introduced to encode spatial features associated with both azimuth and elevation. Then, a lightweight dual-branch network MobileNetV3-DualAngle is developed, which incorporates RG and RB channel modulation blocks to effectively extract cross-channel directional information from the 2D spectrogram representation. Finally, the TriSIS features are fed into the network to estimate both azimuth and elevation angles. Experimental results show that the proposed method achieves high localization accuracy with low computational complexity, showing its potential for practical 3D sound source localization.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

3D Sound Source Localization Based on Four-Element Microphone Array and MobileNetV3-DualAngle Model

  • 期刊论文

  • 发表日期：

    2026年5月4日

查看全文
  • Xiaoguang Wu,
  • Yuxin Qian,
  • Xingyue Cui,
  • Tianwen Guo

摘要

Sound source localization (SSL) plays a fundamental role in spatial audio processing and human–machine interaction. Most existing SSL studies focus on estimating sound source directions in a two-dimensional plane, typically limited to azimuth-only localization. However, in many practical applications such as smart home devices, voice-controlled assistants, and indoor service robots, three-dimensional (3D) directional localization is required. Therefore, this study proposes a sound-intensity-based approach for estimating three-dimensional sound source directions, characterized by azimuth and elevation angles, using a differential microphone array (DMA). Specifically, a Tri-axis Sound Intensity Spectrogram (TriSIS) is first introduced to encode spatial features associated with both azimuth and elevation. Then, a lightweight dual-branch network MobileNetV3-DualAngle is developed, which incorporates RG and RB channel modulation blocks to effectively extract cross-channel directional information from the 2D spectrogram representation. Finally, the TriSIS features are fed into the network to estimate both azimuth and elevation angles. Experimental results show that the proposed method achieves high localization accuracy with low computational complexity, showing its potential for practical 3D sound source localization.