2026-11-13 21:40

Introduction: Capable of understanding incomplete speech, adjusting mid-sentence, invoking tools, and singing songs on demand. What is Hy Realtime? Hy Realtime is Tencent Hunyuan’s real-time voice model designed for authentic conversational interactions.

What is Hy Realtime?

Hy Realtime is Tencent Hunyuan’s real-time voice model tailored for realistic conversations, capable of accurately interpreting ambiguous, fragmented, or corrected utterances in real time. It delivers robust long-form, multi-turn comprehension and instruction adherence. On the expression side, it offers natural audio quality, prosody, and emotional nuance, enabling empathetic communication. It supports on-demand internet search, tool invocation, and song requests based on artist, genre, language, year, and popularity. Hy Realtime has been deeply integrated with Tencent Music’s Lingyin and is already live on Yuanbao voice calls and Tencent customer service, with upcoming API access via Tencent Cloud.

Key Features of Hy Realtime

• Real-Time Speech Understanding: Accurately interprets incomplete inputs such as mumbling, half-sentences, and mid-sentence corrections by leveraging context; maintains stability even in noisy, dynamic call environments.
• Multi-Turn Instruction Following: Continuously manages conversation context during extended dialogues, ensuring accurate execution of complex, sequential, or multi-turn commands.
• Anthropomorphic Spoken Expression: Significantly enhanced audio quality, pronunciation, prosody, and emotional expressiveness enable more natural tone shifts and empathetic interaction.
• Agentic Tool Invocation: Automatically determines whether to perform internet search based on conversational intent, unifying the chain of “decision—retrieval—response generation”.
• On-Demand Song Singing: Supports song requests via combinations of artist, genre, language, year, and popularity, and can continue selecting and singing songs within ongoing context.
• Low-Latency Streaming Generation: Text and voice are jointly modeled, enabling real-time text generation and voice synthesis simultaneously, significantly reducing first-response latency.
• Productized Deployment: Already integrated into Yuanbao voice calls and Tencent customer service; will soon be available via Tencent Cloud API.

How to Use Hy Realtime

• Experience via Yuanbao App
  • Download and update to the latest version of the Yuanbao App.
  • Open Yuanbao, tap the phone icon in the top-right corner to enter voice call mode.
  • In the “Free Conversation” scenario, switch the voice character to “Chuyu”.
  • Begin natural conversation; for song requests, simply say phrases like “Play an English folk song from 2010 to 2020”—tool invocation and internet search will be automatically triggered based on context.
• Web Demo Experience
  • Visit the web demo at https://rtcube.cloud.tencent.com/component/hunyuan/index.html to experience selected capabilities.

Core Advantages of Hy Realtime

• Superior Realistic Speech Comprehension: Accurately handles ambiguous, incomplete, or corrected inputs; maintains robust performance in noisy, variable conversational environments.
• Stable Long-Form Conversations: Enhanced multi-turn instruction following and context management ensure consistent understanding even in prolonged, continuous dialogues.
• More Human-Like Expression: Comprehensive upgrades in audio quality, pronunciation, prosody, and emotional delivery enable more natural emotional progression and empathy.
• Intelligent Tool Invocation: Decides autonomously whether to search online based on context and intent, unifying the retrieval-to-response workflow.
• New Singing Capability: Supports song requests by combining artist, genre, language, year, etc., and can continue selecting and singing songs based on conversation context.
• Low-Latency Streamed Generation: Joint modeling of text and voice enables real-time text generation alongside voice synthesis, minimizing first-response latency.
• Complete Capability Transfer: Cross-modal alignment through pre-training and post-training reduces dilution of knowledge, logic, and reasoning due to “natural language” effects.
• Systematic Training Framework: Multi-stage training pipeline (CT/SFT/MOPD/RL) with dual-reward mechanisms enhances IQ, expression consistency, and task execution coherence.

Competitive Comparison with Similar Models

Application Scenarios of Hy Realtime

• Smart Customer Service & After-Sales Support: Understands user ambiguity, mid-sentence corrections, and emotional cues; performs internet queries when needed and responds naturally.
• Personal Voice Assistant / Companionship: Engages in long-form dialogues with empathetic interaction, managing context for calendar scheduling, Q&A, and daily life services.
• In-Car & Smart Hardware Interaction: Maintains low-latency, stable conversation despite noise, interruptions, and incomplete speech patterns.
• Education & Language Practice: Provides multi-turn error correction, pronunciation/expression feedback, and adjusts tone based on learner’s emotional state.
• Music Entertainment Interaction: Enables on-demand song selection by genre, language, year, and live singing—ideal for companionship, live streaming interaction, and voice-based entertainment entry points.

Source: AI Tools Hub