Scholarlink

This paper presents implementing a Chinese conversational digital avatar system based on NVIDIA ACE (Avatar Cloud Engine). The system enables modular expansion and customization according to diverse application requirements by leveraging the highly flexible microservice architecture of NVIDIA ACE, combined with Kubernetes cluster management. We construct a Chinese speech interaction pipeline on this foundation, integrating modules for automatic speech recognition, natural language understanding, text-to-speech synthesis, and facial animation, thereby achieving a product-level virtual avatar solution. There are different workflows in ACE 4.1 Omniverse Rendering, Unreal Rendering, and the latest 5.0 architecture, and we compare the differences in realism, interactivity, and development flexibility among these rendering technologies during implementation. Experimental results demonstrate that the NVIDIA ACE architecture offers high scalability and stability, allowing seamless integration of third-party Chinese speech and language models to enhance Chinese virtual avatars’ interaction quality and application scope. This study focuses on system integration and productization experience, elaborates on the challenges encountered and solutions adopted during development, and provides a reference for future development and deployment of Chinese conversational avatar applications.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Microservice-Based Implementation of Chinese Conversational Digital Avatars Using NVIDIA ACE

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • FuChe Wu,
  • KuoHsiung Chen,
  • Andrew Dellinger

摘要

This paper presents implementing a Chinese conversational digital avatar system based on NVIDIA ACE (Avatar Cloud Engine). The system enables modular expansion and customization according to diverse application requirements by leveraging the highly flexible microservice architecture of NVIDIA ACE, combined with Kubernetes cluster management. We construct a Chinese speech interaction pipeline on this foundation, integrating modules for automatic speech recognition, natural language understanding, text-to-speech synthesis, and facial animation, thereby achieving a product-level virtual avatar solution. There are different workflows in ACE 4.1 Omniverse Rendering, Unreal Rendering, and the latest 5.0 architecture, and we compare the differences in realism, interactivity, and development flexibility among these rendering technologies during implementation. Experimental results demonstrate that the NVIDIA ACE architecture offers high scalability and stability, allowing seamless integration of third-party Chinese speech and language models to enhance Chinese virtual avatars’ interaction quality and application scope. This study focuses on system integration and productization experience, elaborates on the challenges encountered and solutions adopted during development, and provides a reference for future development and deployment of Chinese conversational avatar applications.