基于MFCC与DTW融合的声纹识别算法在移动支付中的优化研究

Research on the Optimization of Voiceprint Recognition Algorithms Based on the Fusion of MFCC and DTW for Mobile Payment Applications

摘要: 本研究聚焦MFCC与DTW融合的声纹识别算法在移动支付场景中的优化应用，针对移动支付环境噪声复杂、终端硬件性能差异大、用户语音时长不一及安全攻击风险高等核心挑战，系统分析了传统算法在特征鲁棒性、信道适应性、匹配精度与计算效率等方面的瓶颈。在特征提取环节，引入小波阈值降噪与自适应Mel滤波器组，融合多维特征参数，并面向嵌入式平台进行轻量化设计；在匹配环节，提出加权DTW路径约束、模板自适应更新及DTW与GMM两步匹配策略，构建基于冒名者分布的动态门限判决机制。实验结果表明，优化后的算法在移动支付场景下识别准确率显著提升，响应时间控制在3秒以内，显著提升了噪声鲁棒性、跨设备兼容性与防攻击能力。研究成果为移动支付提供了高效安全的声纹认证方案，具有广泛的应用前景。

Abstract: This study focuses on optimizing the integration of MFCC and DTW-based voiceprint recognition algorithms for mobile payment applications. Addressing core challenges in this environment, including complex ambient noise, significant variations in device hardware performance, diverse user speech durations, and high security risks, the research systematically analyzes limitations of traditional algorithms in feature robustness, channel adaptability, matching accuracy, and computational efficiency. During feature extraction, we employ wavelet threshold-based noise reduction and adaptive Mel filter banks to combine multidimensional features while designing a lightweight solution for embedded platforms. For matching, we propose weighted DTW path constraints, template adaptive updates, and a two-step DTW-GMM matching strategy, establishing a dynamic threshold decision mechanism based on impostor distribution patterns. Experimental results demonstrate that the optimized algorithm achieves significantly improved recognition accuracy with response times under 3 seconds, enhanced noise resistance, cross-device compatibility, and stronger attack resilience. These findings provide an efficient and secure voiceprint authentication solution with broad application potential for mobile payments.