arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2409.08552eess.AScs.SD
收藏

统一音频事件检测

Unified Audio Event Detection

National University of Singapore(新加坡国立大学);Shanghai Jiao Tong University(上海交通大学);Alibaba Group(阿里巴巴集团)

Yidi Jiang, Ruijie Tao, Wen Huang, Qian Chen, Wen Wang

AI总结:

针对声音事件检测与说话人分割聚类在复杂音频场景下分析不全面的问题,提出统一音频事件检测任务及基于Transformer的T-UAED框架,构建对应数据集,性能优于简单组合基线且在单任务上表现与专用模型相当。

摘要:

声音事件检测(Sound Event Detection, SED)用于检测声音事件的区域,而说话人分割聚类(Speaker Diarization, SD)则将语音对话按不同说话人进行分段。在SED中,所有说话人片段都被归类为单一语音事件,而在SD中,非语音声音仅被视为背景噪声。因此,在同时包含语音对话和非语音声音的复杂音频场景中,这两项任务都只能提供部分分析。本文提出了一项名为统一音频事件检测(Unified Audio Event Detection, UAED)的新型任务,用于实现全面的音频分析。UAED挖掘SED与SD任务之间的协同作用,能够同时检测非语音声音事件和基于说话人身份的细粒度语音事件。为解决这一任务,我们提出了基于Transformer的UAED(T-UAED)框架,并基于Librispeech数据集和DESED音库构建了UAED Data。实验表明,所提框架有效利用了任务间的交互作用,其性能显著优于简单组合SED和SD模型输出的基线方法。此外,T-UAED还展现出通用性:在DESED和CALLHOME数据集上,其表现与专门针对单一SED和SD任务的专用模型性能相当。

英文摘要:

Sound Event Detection (SED) detects regions of sound events, while Speaker Diarization (SD) segments speech conversations attributed to individual speakers. In SED, all speaker segments are classified as a single speech event, while in SD, non-speech sounds are treated merely as background noise. Thus, both tasks provide only partial analysis in complex audio scenarios involving both speech conversation and non-speech sounds. In this paper, we introduce a novel task called Unified Audio Event Detection (UAED) for comprehensive audio analysis. UAED explores the synergy between SED and SD tasks, simultaneously detecting non-speech sound events and fine-grained speech events based on speaker identities. To tackle this task, we propose a Transformer-based UAED (T-UAED) framework and construct the UAED Data derived from the Librispeech dataset and DESED soundbank. Experiments demonstrate that the proposed framework effectively exploits task interactions and substantially outperforms the baseline that simply combines the outputs of SED and SD models. T-UAED also shows its versatility by performing comparably to specialized models for individual SED and SD tasks on DESED and CALLHOME datasets.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑