Alibaba Qwen Team Releases Qwen3.8-LiveTranslate: A Real-Time Interpretation Model That Cuts Average Lag to 2.3 Seconds Across 60 Languages
https://www.marktechpost.com/2026/09/19/alibaba-qwen-team-releases-qwen3-8-livetranslate/📌 阿里 Qwen 團隊發布同步口譯模型，平均延遲砍到2.3秒
TL;DR：Qwen3.8-LiveTranslate 用新的 Interleave 架構邊聽邊譯，60種語言即時互通，已上線 API。
同步口譯向來是個矛盾的任務：等得越久，模型看到的上下文越完整；講得越快，聽者等待的時間越短。阿里 Qwen 團隊這次交出的答案，是把這個兩難重新設計了一次。
🤔 同步口譯的老問題：等待與延遲的拔河
Qwen 團隊發布 Qwen3.8-LiveTranslate，這是新一代即時同步口譯模型，能一邊聽取即時語音（可選搭配視訊畫面），一邊在說話者仍在講話時就同步輸出翻譯文字與語音。衡量這類系統的關鍵指標是 LAAL（Length-Adaptive Average Lagging），用來計算翻譯內容平均落後原始語音多少時間，同時避免系統靠「過度生成內容」灌水分數。
🧩 核心是一套新的 Interleave 架構
Qwen 團隊表示，此模型是 Qwen3.8-LiveTranslate-Flash 的即時版本，建構在 Qwen-Omni 技術堆疊、大規模多模態資料、跨語言與跨模態對齊，以及視覺增強能力之上。Flash 版本本身也支援離線的音訊與影片翻譯。
模型可理解60種語言，其中29種能以語音加文字形式輸出，其餘31種則僅輸出文字。支援語音輸出的語言包括中文、英文、阿拉伯文、德文、法文、西班牙文、日文、韓文、印地文等。輸入端接受音訊與可選的圖片，官方文件建議每秒最多傳送2張圖片；嘴型、手勢、螢幕上的文字等視覺線索，能在嘈雜環境或語意模糊時提供輔助判斷。此外系統支援「熱詞」設定，可將來源語言的特定詞彙對應到固定的目標翻譯，文件建議熱詞數量上限為1,000個。這代表模型並非單純的語音辨識加翻譯串接，而是把說話者分離（real-time speaker diarization）、雙語同步顯示、長上下文消歧義等能力整合進同一套推理流程。
📊 延遲降1成8，但別忽略配額與價格
Qwen 團隊報告 LAAL 從2.8秒降至2.3秒，約降低18%的平均延遲，同時提升忠實度、流暢度與簡潔度。開發者可透過 WebSocket Realtime API 串接，模型 ID 為 qwen3.8-livetranslate-flash-realtime，預設的分段判斷方式為 speaker_detection，音訊輸入預設為16kHz PCM、輸出為24kHz PCM，預設語音為 Tina。session.output_modalities 可設定為純文字或文字加語音，且務必在關閉連線前送出 session.finish，否則最後一段翻譯內容會遺失。
情境成本方面，音訊輸入每秒消耗7個 token，音訊輸出每秒消耗12.5個 token；一小時的語音輸入加輸出，在新加坡節點約需1.54美元（尚未計入文字與圖片 token）。文件也列出北京節點每百萬 token 的定價分別為5.653、0.466、14.133、22.613美元，較新加坡節點更低。上下文視窗為53,248個 token（輸入49,152、輸出4,096），預設速率限制為每分鐘10次請求、10萬個 token。
⚠️ 目前還不支援的功能
Model Studio 文件明確列出目前不支援 function calling、structured outputs、batch inference 與 fine-tuning，這代表現階段它比較適合做即時口譯這類垂直場景，而非泛用的 agent 後端。
🎯 實務啟示
對於需要跨語言即時溝通的產品（視訊會議、直播字幕、客服系統）而言，這次更新是漸進式但可直接落地的改良：延遲降低、支援視覺線索與熱詞設定，都是實際部署時會在意的細節。若要整合，需留意 token 計費方式與速率限制，並確保正確處理 session 生命週期以避免漏字。
🔗 來源
- 標題：Alibaba Qwen Team Releases Qwen3.8-LiveTranslate: A Real-Time Interpretation Model That Cuts Average Lag to 2.3 Seconds Across 60 Languages
- 作者／機構：Asif Razzaq, MarkTechPost
- 連結：https://www.marktechpost.com/2026/09/19/alibaba-qwen-team-releases-qwen3-8-livetranslate/
#Qwen #Alibaba #RealTimeTranslation #SimultaneousInterpretation #SpeechAI #MultimodalAI #WebSocketAPI #LLM #QwenOmni #AIProductRelease
原始資料 MarkTechPost · 收集於 2026-09-20
摘要原文
Qwen has released Qwen3.8-LiveTranslate , its next-generation real-time simultaneous interpretation model. It listens to live speech, with optional video frames, and returns translated text and speech while the speaker is still talking. The core change is a new Interleave architecture. Qwen reports gains in faithfulness, fluency, and conciseness, with average lagging (LAAL) dropping from 2.8 seconds to 2.3 seconds. The release also adds real-time speaker diarization, synchronized bilingual display, and long-context disambiguation. Deployable? Yes, as a hosted API. It is live on Alibaba Cloud Model Studio and QwenCloud as qwen3.8-livetranslate-flash-realtime over WebSocket. Simultaneous interpretation is a tradeoff. Waiting longer gives the model more context. Speaking sooner cuts delay for the listener. Qwen3.8-LiveTranslate rebuilds this loop with an Interleave architecture. The latency metric here is LAAL, or Length-Adaptive Average Lagging . It measures how far the translation trails the source speech on average. It also avoids rewarding systems that over-generate output. A drop from 2.8 seconds to 2.3 seconds is roughly an 18% cut in average lag. QwenCloud team describes the model as the real-time version of Qwen3.8-LiveTranslate-Flash. It builds on the Qwen-Omni stack, large-scale multimodal data, cross-language and cross-modal alignment, and visual enhancement. The Flash model also supports offline audio and video translation. Try the explainer below. It walks through the interleaved stream, speaker tagging, context disambiguation, language coverage, and session cost. The model understands 60 languages. It can speak 29 of them, returning audio plus text. The remaining 31 return text only. Speech output covers Chinese, English, Arabic, German, French, Spanish, Japanese, Korean, Hindi, and others. Inputs are audio and optional images. Outputs are text and audio. Visual cues such as lip movements, gestures, and on-screen text help in noisy rooms and with ambiguous words. The docs recommend sending no more than 2 images per second. Teams can also set hotwords. These map source terms to fixed target translations. The docs recommend configuring no more than 1,000 hotwords. Developers connect through the WebSocket Realtime API with the model ID qwen3.8-livetranslate-flash-realtime . The default turn detection type is speaker_detection . Clients stream audio continuously and receive server-generated responses. Default audio is 16 kHz PCM in and 24 kHz PCM out. The default voice is Tina. Set session.output_modalities to text only, or text and audio. Always send session.finish before closing, or the final segment is lost. Singapore list pricing per 1M tokens: Beijing pricing is lower, at $5.653, $0.466, $14.133, and $22.613 in USD. Audio input consumes 7 tokens per second. Audio output consumes 12.5 tokens per second. One hour of speech in and speech out costs about $1.54 in Singapore, before text and image tokens. The context window is 53,248 tokens, with 49,152 for input and 4,096 for output. Default rate limits are 10 requests and 100,000 tokens per minute. Model Studio lists function calling, structured outputs, batch inference, and fine-tuning as unsupported. Check out the Technical Details . All credit goes to the researcher of this project. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter . Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us The post Alibaba Qwen Team Releases Qwen3.8-LiveTranslate: A Real-Time Interpretation Model That Cuts Average Lag to 2.3 Seconds Across 60 Languages appeared first on MarkTechPost .
由 claude-code/sonnet 自動生成