利用子带多假设回声消除器统计特征，实现高精度声学场景分析。
Sub-band Domain Multi-Hypothesis Acoustic Echo Canceler Based Acoustic Scene Analysis
- 基于子带多假设回声消除器提取多维统计特征
- 在双讲、回声路径变化等复杂场景下表现稳定
- 无需额外计算开销，适配现有回声消除系统
本文提出一种新型声学场景分析方法，通过从子带域多假设回声消除器（SDMH-AEC）中提取一组统计特征实现。该设计采用多种自适应滤波策略，在收敛、扰动和稳态条件下具有潜在互补行为。通过跨子带聚合统计量，构建出具有强判别能力的特征向量，可用于区分不同声学事件并估计声学参数。由于各滤波器间的互补性，该方法在几乎无额外计算成本的情况下，提供了丰富的信息源。实验使用包含双讲、回声路径变化及全双工设备物理移动的真实数据验证了其有效性。所提取特征可直接用于现有回声消除算法与技术中的声学场景分析任务。
原文摘要 · Abstract (English)
This paper introduces a novel approach for acoustic scene analysis by exploiting an ensemble of statistics extracted from a sub-band domain multi-hypothesis acoustic echo canceler (SDMH-AEC). A well-designed SDMH-AEC employs multiple adaptive filtering strategies with potentially complementary behaviours during convergence, perturbations, and steady-state conditions. By aggregating statistics across the sub-bands, we derive a feature vector that exhibits strong discriminative power for distinguishing different acoustic events and estimating acoustic parameters. The complementary nature of the SDMH-AEC filters provides a rich source of information that can be extracted at insignificant cost for acoustic scene analysis tasks. We demonstrate the effectiveness of the proposed approach experimentally with real data containing double-talk, echo path change and events where the full-duplex device is physically moved. The extracted features enable acoustic scene analysis using existing echo cancellation algorithms and techniques.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。