Qwen3.5-LiveTranslate - A real-time simultaneous interpretation model launched by Alitongyi

What is Qwen3.5-LiveTranslate?

Qwen3.5-LiveTranslate is a large-scale real-time simultaneous interpretation model launched by the Alibaba Tongyi team. It supports input in 60 languages, output in 29 languages, and 3500+ translation combinations. Through readable unit streaming technology, it compresses the end-to-end word latency to 2.8 seconds. The model has real-time timbre cloning and hot word enhancement capabilities and is designed for cross-border meetings, live streaming overseas, and global collaboration scenarios.

The main functions of Qwen3.5-LiveTranslate

• Multilingual full-coverage translationIt supports listening, speaking, reading and writing in 60 languages and voice output in 29 languages, forming 3500+ translation pairs, covering mainstream business and high-frequency translation scenarios worldwide.
• Ultra-low latency streaming simultaneous interpretationBased on the new "readable unit" technology, it achieves streaming output with an end-to-end word latency as low as 2.8 seconds, meeting the requirements for smooth live streaming and real-time meetings.
• Real-time timbre cloningWith just one sentence, a user's original voice can be cloned, maintaining the consistency of voice quality and identity when outputting across languages.
• Enhanced accuracy of hot wordsIt supports dynamic configuration and priority recognition of personal names, brands, and industry terms to ensure accurate translation of terms in professional scenarios.

The technical principle of Qwen3.5-LiveTranslate

• Streaming readable unit architectureThe voice input is segmented into the smallest semantic units for real-time encoding and decoding, avoiding the traditional batch processing mode of waiting for complete sentences and significantly reducing end-to-end latency.
• End-to-end speech large modelBased on the Qwen3.5 multimodal platform, it unifies the modeling of speech recognition, translation and speech synthesis, reduces cascading errors and improves the naturalness of cross-language speech conversion.
• Timbre Consistency CodingBy extracting speaker features and reconstructing vocoders, the original timbre, intonation, and emotional features are preserved in the translated target language speech.
• Hot word dynamic injection mechanismInjecting a user-defined terminology library into the model context before inference improves the recognition priority and translation accuracy of specific words.

How to use Qwen3.5-LiveTranslate

• Access Experience EntryOpen your browser and access the QwenChat Qwen3.5-LiveTranslate model entry point: https://omni.qwen.ai/ or wait for the Alibaba Cloud Bailian platform to launch.
• Select simultaneous interpretation modeSwitch to LiveTranslate real-time translation mode in the dialog interface and set the source language and target language.
• Configure a hot word library (optional)Upload or enter technical terms, names of people and brands involved in the meeting/live broadcast to improve translation accuracy.
• Enable real-time simultaneous interpretationClick the start button, speak into the microphone, and the system will output the translated target language audio in real time.
• Tone cloning settings (optional)Record a sample of your voice and enable the "Use My Voice" feature to keep the original voice tone in the translation output.

The core advantages of Qwen3.5-LiveTranslate

• Industry-leading latencyThe average word delay of 2.8 seconds is close to the level of human simultaneous interpretation, which is far superior to traditional cascaded solutions.
• High sound fidelityReal-time cloning technology makes cross-language communication more personalized and approachable.
• Professional scenario adaptationThe hot word enhancement feature addresses the pain points of translating brand names and technical terms.