两篇新的研究论文介绍了改进自动语音识别（ASR）系统的方法。第一篇论文“MURMUR”提出了一个高效的推理系统，专为长篇幅ASR设计，通过处理中间大小的音频块和优化注意力稀疏性来平衡准确性和低延迟。第二篇论文“SN-WER”提出了一种名为脚本归一化的词错误率（SN-WER）的新评估指标，通过在比较前对脚本进行归一化，来解决标准词错误率计算在处理多语言ASR（特别是印度语言）时出现的准确性问题。 AI
影响 这些论文引入了提高ASR准确性和评估的新技术，有可能为不同语言和长篇幅内容带来更强大的语音转文本系统。
排序理由 两篇在arXiv上发表的学术论文，介绍了新的ASR方法。
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
arXiv:2606.01483v1 Announce Type: cross Abstract: Long-form automatic speech recognition (ASR) requires both high accuracy and low latency, but existing systems force a trade-off between the two. Chunk-based pipelines process audio in parallel windows for low latency, but lose cr…
arXiv:2606.02548v1 Announce Type: new Abstract: Word Error Rate (WER) is the dominant metric for automatic speech recognition (ASR), but it can overestimate errors when references and hypotheses encode the same words in different scripts. This issue is common in multilingual sett…
Word Error Rate (WER) is the dominant metric for automatic speech recognition (ASR), but it can overestimate errors when references and hypotheses encode the same words in different scripts. This issue is common in multilingual settings where ASR models may emit romanized text. W…