北京测绘2026，Vol.40Issue(6)：787-794,8.DOI:10.19580/j.cnki.1007-3000.2026040093
车载导航大模型语音提示生成方法
Voice prompt generation method based on large language model for in-car navigation
摘要
Abstract
In-car navigation systems require voice prompts to be both precise and context-aware,but traditional rule-based methods often struggle in dynamic traffic scenarios.For the application of large language models in in-car navigation,this paper proposed a voice prompt generation(VPG)method,which solved several key technical challenges while generating structured voice instructions.VPG utilized a template selector to retrieve high-quality templates,thereby ensuring the con-trollability of output and compliance with the required structure;it utilized a keyphrase predictor to extract key information from structured navigation data to strike a balance between accuracy and completeness and generated fluent and context-relevant prompts through a sequence generator.This method could integrate static navigation information and real-time traf-fic multi-source data to meet the real-time processing requirements of in-car scenarios.Experiments on large-scale real datas-ets show that VPG outperforms traditional methods in accuracy,fluency,and conciseness and can provide effective technical support for the voice prompt generation of large language models in in-car navigation.关键词
车载导航/语音提示/数据到文本/大语言模型/模板检索/关键信息预测Key words
in-car navigation/voice prompt/data-to-text/large language model/template retrieval/key information prediction分类
天文与地球科学引用本文复制引用
冯硕,冯泽邦,郭静桐,冯仲科..车载导航大模型语音提示生成方法[J].北京测绘,2026,40(6):787-794,8.基金项目
北京市自然科学基金(8232038) （8232038）
北京林业大学5.5工程科研创新团队项目(BLRC2023A03) （BLRC2023A03）
北京林业大学科技创新计划(XJJSKYQD202632). （XJJSKYQD202632）
