一种新的车载语音信号对齐框架

打开文本图片集

中图分类号：U463.6 文献标识码：A 文章编号：1003-8639（2026）03-0082-02

A New Framework for Vehicle-mounted Voice Signal Alignment

Li Fengbo， Shi Xiaoyong， Wang Zhisheng， Hao Jingyang， Zhao Yinqi， Yin Yiting （Shijiazhuang Institute of Railway Technology，Shijiazhuang O5oO41， China）

【Abstract】Dynamic Time Warping （DTW） is commonly used for smal-vocabulary speech recognition in invehicle voiceinteraction，butthealgorithmhas fixed parametersand struggles tocope withspeechrate variationsand noise interference.This paper proposes an Adaptive DTW （A-DTW）framework，validated through experiments on an isolated word corpus with variable speedsand multiple types of noise.The results show that its recognition accuracy in complex scenarios is3.5～8.7percentagepoints higherthanthatofthe traditionalDTWanditsmainstreamimprovedalgorithms，with manageablecomputationaloverhead，providinganefectivesolutionforstable speechrecognition inresource-constrained environments.

【Key Words】 in-vehicle speech recognition;Dynamic Time Warping;adaptive algorithm;robustness; feature fusion

0 引言

车载语音交互系统受车载端硬件资源约束，小词汇量孤立词语音识别任务中，动态时间规整（DTW）因模型简单、无需大量训练数据，成为该领域实用且具代表性的方法[1]。（剩余2204字）