基于大语言模型的银行智能语音助手

Bank Intelligent Voice Assistant Based on Large Language Models

摘要: 针对传统银行智能语音助手存在的语义理解僵化、上下文关联能力弱及业务覆盖范围有限等问题，本文提出一种基于大语言模型(LLM)的新一代智能语音助手架构。该架构以本地化部署的轻量级LLM为核心处理引擎，通过模型量化技术保障其在普通计算设备上的高效推理；引入检索增强生成(RAG)技术构建银行业务知识库，用以增强对复杂金融咨询的准确响应能力。以所提框架为基础，设计并实现了以DeepSeek-R1 (7B)模型为核心、文本和语音交互为媒介的智能银行语音客服系统。实验结果表明，在典型银行业务场景中，该系统能精准理解用户意图并完成账户查询、转账等任务，展现出良好的实用性与适应性。

Abstract: Aiming at the problems existing in the traditional bank intelligent voice assistant, such as rigid semantic understanding, weak context correlation ability and limited business coverage, this paper proposes a new generation of intelligent voice assistant architecture based on large language model (LLM). The architecture takes the lightweight LLM deployed locally as the core processing engine, and ensures its efficient reasoning on ordinary computing devices through model quantification technology. The retrieval enhanced generation (RAG) technology is introduced to build a banking knowledge base to enhance the accurate response ability to complex financial consulting. Based on the proposed framework, an intelligent bank voice customer service system with DeepSeek-R1 (7B) model as the core and text and voice interaction as the media is designed and implemented. The experimental results show that in typical banking business scenarios, the system can accurately understand the user’s intention and complete the tasks of account query and transfer, showing good practicability and adaptability.