English(EN) Spoken Language Identification with Pre-trained Models and Margin Loss

研究人员开发使用预训练模型和边界损失的新语音语言识别方法

作者 PulseAugur 编辑部 · [2 个来源] · 2026-05-03 14:37

研究人员开发了一种使用预训练模型和基于边界的损失进行语音语言识别的新方法。该方法增强了语言表示区分不同语言的能力，同时最大限度地减少了说话人特征的影响。在Tidy-X数据集上的实验表明，与基线相比，宏观准确率提高了45.7%，微观准确率提高了15.2%，均有显著提升。 AI

影响提高了语音语言识别的准确性，可能有助于多语言应用和转录服务。

排序理由这是一篇详细介绍语音语言识别新方法的学术论文。

AI 生成摘要 · Google Gemini · 来自 2 个来源。我们如何撰写摘要 →

报道来源 [2]

arXiv cs.CL TIER_1 English(EN) · Zhihua Fang, Liang He, Weiwu Jiang · 2026-05-05 04:00

Spoken Language Identification with Pre-trained Models and Margin Loss

arXiv:2605.01905v1 Announce Type: cross Abstract: For the speaker-controlled spoken language identification task proposed in the TidyLang Challenge 2026, this paper proposes a language identification method based on pre-trained models and margin-based losses. The proposed method …
arXiv cs.CL TIER_1 English(EN) · Weiwu Jiang · 2026-05-03 14:37

Spoken Language Identification with Pre-trained Models and Margin Loss

For the speaker-controlled spoken language identification task proposed in the TidyLang Challenge 2026, this paper proposes a language identification method based on pre-trained models and margin-based losses. The proposed method adopts a pre-trained ECAPA-TDNN as the feature enc…