两篇新研究论文提出了区分中文方言的先进方法，这项任务由于文本数据有限而历来具有挑战性。一篇论文介绍了一种语音驱动的方法，使用梅尔频率倒谱系数（MFCC）和CNN-HMM-DNN模型从音频中识别方言。第二篇论文侧重于低资源场景，采用迁移学习和数据增强技术，并结合CDDTLDA框架和自注意力机制，以改进方言鉴别的自动语音识别。 AI
影响 这些论文在细粒度语言分析方面提升了AI能力，有望改善面向不同语言社区的语音识别和自然语言处理应用。
排序理由 两篇在arXiv上发表的学术论文，详细介绍了语言鉴别的新方法。
AI 生成摘要 · Google Gemini · 来自 4 个来源。 我们如何撰写摘要 →
arXiv:2606.18584v1 Announce Type: new Abstract: Language discrimination among similar languages, varieties, and dialects is a challenging natural language processing task. The traditional text-driven focus leads to poor results. In this paper, we explore the effectiveness of spee…
arXiv:2606.18597v1 Announce Type: new Abstract: Chinese dialects discrimination is a challenging natural language processing task due to scarce annotation resource. In this article, we develop a novel Chinese dialects discrimination framework with transfer learning and data augme…
Chinese dialects discrimination is a challenging natural language processing task due to scarce annotation resource. In this article, we develop a novel Chinese dialects discrimination framework with transfer learning and data augmentation (CDDTLDA) in order to overcome the short…
Language discrimination among similar languages, varieties, and dialects is a challenging natural language processing task. The traditional text-driven focus leads to poor results. In this paper, we explore the effectiveness of speech-driven features towards language discriminati…