整页是一条对谈稿:每句一块,页边是 Speaker 和时间;上面小字是原文(注释),下面大字是译文(正文)。底部直播条:原声 = 实时识别,快译 = 草稿翻译,句子落定后被正式译文覆盖。
A single bilingual script: each turn shows the speaker and time in the margin, the original in small type (annotation) above the translation in large type (the text). The live strip at the bottom carries live recognition and the rough draft, replaced by the refined translation at sentence end.
顶栏"显示"选的是你阅读的语言:选中文或 English 后隐藏原文行,和你语言相同的句子直接显示原文,不同的显示译文。
The view selector picks YOUR reading language: originals hide, sentences in your language show as-is, the rest show their translation.
支持中↔EN、EN↔VI、中↔VI。中英模式有逐字快译、热词和说话人分离；含越南语的模式按整句出字，无热词和说话人分离。识别纠错和翻译术语表在三种模式中都照常使用。
Supports ZH↔EN, EN↔VI, and ZH↔VI. ZH↔EN includes live drafts, hotwords, and speaker separation. Vietnamese pairs caption whole sentences without hotwords or speaker separation. Recognition corrections and the translation glossary remain active in all three modes.
识别:豆包流式语音识别 2.0(二遍识别、热词增强、说话人聚类)。翻译:火山引擎机器翻译大模型,术语表逐句注入。
ASR: Volcengine Seed-ASR 2.0 (two-pass, hotwords, speaker clustering). Translation: Volcengine MT LLM with per-sentence glossary.
音频流送往火山引擎做识别和翻译,本地不保存音频;字幕文本只存在这个页面和你导出的文件里。
Audio is processed by Volcengine cloud APIs and never stored locally; captions live only in this page and your exported files.
Repo: kydchen/CKB_mBabel
实验控制仅主持人本机可用。Lab controls are only available on the host machine.
精翻默认带前两句上下文(ark),术语表照常注入;无独立润色开关。
The refined pass carries sentence context by default (ark); glossary still applies. No separate polish toggle.
近 60 秒请求 Requests 60s:…
修的是语音识别听错的词(原文里的错字),不是翻译。写法:错词=正确词,例:CKG=CKB。即时对后续句子生效,不回改历史;退出时存入术语表。 This fixes what the ASR misheard (wrong words in the ORIGINAL line), not translation. Format: wrong=right, e.g. CKG=CKB. Applies to future sentences only; persists on exit for the next meeting.
点名字,主屏会跳到并高亮 TA 说过的句子,方便确认是谁。浮窗不会关闭。
Click a name to jump to and highlight their lines behind this dialog.
本次会议还没有识别到说话人。No speakers detected yet.
输过的名字会记在本机浏览器,下次点输入框直接下拉选择;不会按编号自动填,因为编号和人不绑定。命名后页面上的 Speaker 标签也会换成名字。
Typed names are remembered locally for quick selection and update the speaker labels on the page. Names are never auto-assigned to a cluster number.
删除只会移除下次可选的名字,不会改写本场字幕。
Deleting a saved name only removes the suggestion; it does not change this meeting.