基于距离的单通道移动 GPU 环境声源分离
音频与语音处理
2025-01-07 v1 人工智能
摘要
本研究强调了在户外环境中探索基于距离的声源分离（DSS）的重要性。不同于现有主要关注室内环境的研究，所提出的模型旨在捕捉户外音频源的独特特征。它包含先进的技术，包括两阶段 conformer 块、线性关系感知自注意（RSA）以及 TensorFlow Lite GPU 代理。虽然线性 RSA 可能不像二次 RSA 那样明确捕捉物理线索，但线性 RSA 增强了模型的情境感知能力，导致在户外和室内环境中改进的 DSS 性能。实验结果表明，所提出的模型克服了现有方法的局限性，并在移动设备上显著提高了能源效率和实时推理速度。
引用
@article{arxiv.2501.03045,
title = {Single-Channel Distance-Based Source Separation for Mobile GPU in Outdoor and Indoor Environments},
author = {Hanbin Bae and Byungjun Kang and Jiwon Kim and Jaeyong Hwang and Hosang Sung and Hoon-Young Cho},
journal= {arXiv preprint arXiv:2501.03045},
year = {2025}
}
备注
Accepted by ICASSP2025. \c{opyright} 2025 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component