Abstract: It is difficult to control voice of intelligent vehicles due to vehicle interior noise, which can decrease detection rate of voice obviously. The article presents a detection method based on Hilbert-Huang Transform（HHT） spectral matrix. The method with frame as a unit processes time, frequency, and amplitude matrix of input signal HHT, and constructs amplitude-time curve by analyzing characteristics of HHT and noise, voice signal amplitude distribution. By estimating front end noise, the method sets threshold automatically for detecting phonetic segment of the whole signal. The result indicates that the method can detect voice activity effectively with vehicle interior heavy noise.

Key words: signal processing, Hilbert-Huang Transform（HHT）, vehicle interior noise, signal detection, low signal-
to-noise ratios

摘要： 由于车内噪声的存在，使语音的检测率降低得非常明显，这给智能汽车的语音控制带来了困难。提出了一种基于HHT谱矩阵的检测方法。该方法通过分析HHT的特点，以及噪声、语音信号的幅度分布特点，以帧为单位对输入信号HHT的时、频、幅矩阵进行处理，构建幅值-时间曲线，通过对前端噪声段的估计，自动设定阈值对整个信号的语音段进行检测。实验结果表明，该方法在车内噪声较强的情况下仍能有效检测语音段。

关键词: 信号处理, 希尔伯特-黄变换, 车内噪声, 信号检测, 低信噪比