贝弗尔对话系统实现低延迟语音翻译，中英德三语表现优异。
BeaverTalk: Oregon State University's IWSLT 2025 Simultaneous Speech Translation System
- 分段识别+大模型翻译，用低秩适配微调生成上下文记忆。
- 德语翻译达BLEU 27.83，中文达37.23，延迟低于3.5秒。
- 适合实时会议、直播字幕等低延迟场景使用。
本文介绍贝弗尔对话系统（BeaverTalk）的构建、微调与部署，该系统是参与IWSLT 2025并行语音翻译任务的级联式语音到文本翻译系统。系统架构包括基于语音活动检测（VAD）的分段器、Whisper Large V2用于自动语音识别（ASR），以及Gemma 3 12B作为并行翻译大模型。针对并行翻译任务，采用低秩适配（LoRAs）对大模型进行微调，结合单句源语言记忆库的对话式提示策略以增强上下文理解。系统在英语→德语和英语→中文两种方向上，分别在低延迟与高延迟条件下进行了测试。在英语→德语任务中，系统在流式平均滞后时间（StreamLAAL）分别为1837.86和3343.73时，获得BLEU分数24.64和27.83；在英语→中文任务中，对应流式平均滞后时间为2216.99和3521.35时，取得BLEU 34.07和37.23。
原文摘要 · Abstract (English)
This paper discusses the construction, fine-tuning, and deployment of BeaverTalk, a cascaded system for speech-to-text translation as part of the IWSLT 2025 simultaneous translation task. The system architecture employs a VAD segmenter for breaking a speech stream into segments, Whisper Large V2 for automatic speech recognition (ASR), and Gemma 3 12B for simultaneous translation. Regarding the simultaneous translation LLM, it is fine-tuned via low-rank adaptors (LoRAs) for a conversational prompting strategy that leverages a single prior-sentence memory bank from the source language as context. The cascaded system participated in the English$\rightarrow$German and English$\rightarrow$Chinese language directions for both the low and high latency regimes. In particular, on the English$\rightarrow$German task, the system achieves a BLEU of 24.64 and 27.83 at a StreamLAAL of 1837.86 and 3343.73, respectively. Then, on the English$\rightarrow$Chinese task, the system achieves a BLEU of 34.07 and 37.23 at a StreamLAAL of 2216.99 and 3521.35, respectively.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。