Qwen3.5-LiveTranslate - Alibaba Tongyi's Real-Time Simultaneous Interpretation Model

What is Qwen3.5-LiveTranslate

Qwen3.5-LiveTranslate is a real-time simultaneous interpretation large model launched by the Alibaba Tongyi team. It supports 60 input languages, 29 output languages, and 3,500+ translation combinations. Through readable-unit streaming technology, it compresses the average per-character end-to-end latency to 2.8 seconds. The model features real-time voice cloning and hotword enhancement capabilities, and is purpose-built for cross-border conferences, live streaming going global, and global collaboration scenarios.

Main Features of Qwen3.5-LiveTranslate

• Full Multilingual Coverage Translation: Supports listening, speaking, reading, and writing in 60 languages and speech output in 29 languages, forming 3,500+ translation pairs and covering mainstream global business and high-frequency mutual translation scenarios.
• Ultra-Low-Latency Streaming Simultaneous Interpretation: Based on the all-new "readable unit" technology for streaming output, the average per-character end-to-end latency is as low as 2.8 seconds, meeting the lag-free requirements of live co-streaming and real-time meetings.
• Real-Time Voice Cloning: A single sentence is enough to clone the user's original voice timbre, maintaining vocal texture and identity consistency during cross-language output.
• Precise Hotword Enhancement: Supports dynamic configuration and priority recognition of personal names, brands, and industry terminology, ensuring accurate terminology translation in professional scenarios.

Technical Principles of Qwen3.5-LiveTranslate

• Streaming Readable Unit Architecture: Splits speech input into minimal semantic units for real-time encoding and decoding, avoiding the traditional batch processing mode that waits for complete sentences and significantly reducing end-to-end latency.
• End-to-End Speech Large Model: Based on the Qwen3.5 multimodal foundation, it uniformly models speech recognition, translation, and speech synthesis, reducing cascading errors and improving the naturalness of cross-language speech conversion.
• Voice Timbre Consistency Encoding: Through speaker feature extraction and vocoder reconstruction, it preserves the original timbre, intonation, and emotional characteristics in the translated target-language speech.
• Dynamic Hotword Injection Mechanism: Injects the user-defined terminology library into the model context before inference, improving the recognition priority and translation accuracy of specific vocabulary.

How to Use Qwen3.5-LiveTranslate

• Access the Experience Entry Point: Open a browser and enter the QwenChat Qwen3.5-LiveTranslate model entryhttps://omni.qwen.ai/ or wait forAlibaba Cloud Model Studioplatform launch.
• Select Simultaneous Interpretation Mode: Switch to LiveTranslate real-time translation mode in the chat interface and set the source language and target language.
• Configure Hotword Library (Optional): Upload or enter the professional terminology, personal names, and brand names involved in the meeting/live stream to improve translation accuracy.
• Start Real-Time Simultaneous Interpretation: Click the start button and speak into the microphone, and the system will output translated target-language speech in real time.
• Voice Cloning Settings (Optional): Record a one-sentence voice sample and enable the "Use My Voice" feature so that the translated output maintains the original voice timbre.

Core Advantages of Qwen3.5-LiveTranslate

• Industry-Leading Latency: The 2.8-second average per-character latency is close to the level of human simultaneous interpretation and far superior to traditional cascaded solutions.
• High Voice Timbre Fidelity: Real-time cloning technology makes cross-language communication more personally recognizable and approachable.
• Professional Scenario Adaptability: The hotword enhancement feature addresses the translation pain points of brand names and technical terms.