领域内 SSL 预训练与流式 ASR
In-domain SSL pre-training and streaming ASR
AI总结：
本研究提出在 ATC 环境下利用 4.5k 小时无标签数据进行领域内 SSL 预训练 BEST-RQ 模型，结合分块注意力与动态卷积实现低延迟流式 ASR，显著降低了词错率。
摘要：
在本研究中，我们探讨了领域特定的自监督预训练在空中交通管制（ATC）环境下的离线和流式 ASR 中的优势。我们在 4.5k 小时的无标签 ATC 数据上训练 BEST-RQ 模型，随后在较小的有监督 ATC 数据集上进行微调。为实现实时处理，我们提出使用分块注意力和动态卷积，以确保低延迟推理。我们将这些领域内 SSL 模型与最先进的通用语音编码器（如 w2v-BERT 2.0 和 HuBERT）进行了比较。结果表明，领域适应的预训练在标准 ATC 基准上显著提升了性能，与在广泛语音语料库上训练的模型相比，大幅降低了词错率。此外，所提出的流式方法在更严格的延迟约束下进一步降低了词错率，使其特别适用于安全关键的航空应用。这些发现表明，针对 ATC 数据专门化 SSL 表示是在真实世界运营环境中构建更准确、更高效 ASR 系统的实用途径。
英文摘要：
In this study, we investigate the benefits of domain-specific self-supervised pre-training for both offline and streaming ASR in Air Traffic Control (ATC) environments. We train BEST-RQ models on 4.5k hours of unlabeled ATC data, then fine-tune on a smaller supervised ATC set. To enable real-time processing, we propose using chunked attention and dynamic convolutions, ensuring low-latency inference. We compare these in-domain SSL models against state-of-the-art, general-purpose speech encoders such as w2v-BERT 2.0 and HuBERT. Results show that domain-adapted pre-training substantially improves performance on standard ATC benchmarks, significantly reducing word error rates when compared to models trained on broad speech corpora. Furthermore, the proposed streaming approach further improves word error rate under tighter latency constraints, making it particularly suitable for safety-critical aviation applications. These findings highlight that specializing SSL representations for ATC data is a practical path toward more accurate and efficient ASR systems in real-world operational settings.
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。
