基于C_0复杂度的语音端点检测技术研究

Application of C_0 Complexity Measure in Detecting Speech

复杂性测度的语音端点检测技术,与目前被广泛研究的短时能量、过零率、谱熵以及倒谱等技术相比较,它具有非线性技术的本质特性。实验结果表明C0复杂性测度技术可以较好地实现在动态噪声环境下对语音端点的检测。此技术的实现将有助于提高孤立字语音识别的准确率,同时也将极大的降低语音处理的计算量和复杂性。

This paper proposed a new speech detection method based on complexity measures. It has nonlinear properties, comparing with those generally researched methods, including short-time energy, ZCR, frequency-entropy, LPCC and MFCC. Through the result of a number of experiments, we have found that Co complexity can be a valid feature to make speech/non-speech decision under dynamical noise situation. This new method will improve the accuracy of isolate-word speech recognition, and can also decrease the computations and complexity of speech process.