Scholarlink

<p>Microphone array beamforming is a fundamental technique for speech enhancement. A practical implementation method for microphone array beamforming involves splitting the beamforming filters into two sequential components: first a fixed beamformer, followed by a post-filter gain. Conventional post-filtering methods rely on single-channel noise reduction, assuming that residual noise and target speech are uncorrelated. However, these methods often face challenges in suppressing residual interference, as they lack spatial selectivity and may inadvertently amplify interfering signals. To address this limitation, this paper proposes a neural direction-gain for post-filtering, which is cascaded with a robust superdirective beamformer to enhance directivity and speech quality. The core of the proposed method is a direction-aware post-filter network, in which a mask decoder dynamically estimates the directional gain, while a complex decoder adaptively refines the real and imaginary components of the output signal. This architecture enables joint optimization with the superdirective beamformer, resulting in enhanced interference suppression and improved perceptual quality.</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

Design of Neural Directional Gain Post-filtering for Microphone Array Beamforming

  • 期刊论文

  • 发表日期：

    2026年2月20日

查看全文
  • Yongyi Deng,
  • Xiaoran Yang,
  • Kunlong Zhao,
  • Yanxin Tian,
  • Gongping Huang

摘要

Microphone array beamforming is a fundamental technique for speech enhancement. A practical implementation method for microphone array beamforming involves splitting the beamforming filters into two sequential components: first a fixed beamformer, followed by a post-filter gain. Conventional post-filtering methods rely on single-channel noise reduction, assuming that residual noise and target speech are uncorrelated. However, these methods often face challenges in suppressing residual interference, as they lack spatial selectivity and may inadvertently amplify interfering signals. To address this limitation, this paper proposes a neural direction-gain for post-filtering, which is cascaded with a robust superdirective beamformer to enhance directivity and speech quality. The core of the proposed method is a direction-aware post-filter network, in which a mask decoder dynamically estimates the directional gain, while a complex decoder adaptively refines the real and imaginary components of the output signal. This architecture enables joint optimization with the superdirective beamformer, resulting in enhanced interference suppression and improved perceptual quality.