研究人员评估了 Mamba 架构在七种南非语言中的自动语音识别（ASR）能力，并将其性能与 Conformer 基线进行了比较。Mamba 在准确性方面与 Conformer 相当，同时所需的计算资源更少，训练速度更快。使用 Mamba 进行多语种训练比单语方法提高了性能，尽管显式的语言信息并未提高领域内准确性，但确实增强了跨语料库的鲁棒性。在低资源环境下的消融研究表明，语言嵌入是有益的，它们充当特定任务的控制向量，而不是捕捉语言相似性。 AI
影响 Mamba 在高效且有效的多语种自动语音识别方面显示出潜力，尤其是在资源匮乏的语言环境中。
排序理由 学术论文，详细介绍了模型在特定语言和任务上的评估。[lever_c_demoted from research: ic=1 ai=1.0]
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2607.01502v1 Announce Type: new Abstract: Recent advances in automatic speech recognition (ASR) have explored different sequence models, including Conformer-based models and newer state space models such as Mamba. Although prior work has evaluated these architectures in mul…
Recent advances in automatic speech recognition (ASR) have explored different sequence models, including Conformer-based models and newer state space models such as Mamba. Although prior work has evaluated these architectures in multiple languages, their effectiveness in African …