摘要
在VC++的编程环境下设计了一个小词汇量孤立词非特定人的汽车语音识别系统,系统中的识别词汇都是汽车驾驶员在驾驶过程中可能做的一些手控操作。语音预处理采用的是改进后的端点检测算法。实验分析时,在测试数据中加入噪声库(Noise X-92)中的车内噪声来模拟汽车驾驶环境,并提出随机映射梅尔频率倒谱系数来增强噪声环境下系统的鲁棒性。测试数据表明,使用随机映射特征参数使得系统的抗噪声能力得到了很大的提升。
A speaker-independent speech recognition system is designed with small vocabulary and isolated words by VC++, which can recognize some words that are all manual operations achieved by motorists during the driving and pre-process speech signal by improved endpoint detection algorithm.The vehicle interior noise is added into the test data to simulate the driving environment and put forward Stochastic Mapping Mel Frequency Cepstrum Coefficient ( SM MFCC) to strengthen the robustness of the recognition system.The great advantage of SM_MFCC is proved by experiments.
出处
《电声技术》
2016年第8期36-39,共4页
Audio Engineering
基金
国家自然科学基金项目(61201213)
莙政基金项目(21315003)