摘要
为提高低信噪比时带噪语音端点检测的性能,提出了一种将扩展的谱相减法与SAP(Speech AbsenceProbab ility)软门限相结合的方法。采用基于噪声补偿结构的扩展谱相减法,通过使用自适应的判决规则,在不需要进行语音激活检测的情况下有效地去除了背景噪声,克服了单麦克输入时无法在语音段对噪声进行估计的缺点。同时采用非语音段概率SAP软门限,直接对增强后的语音信号进行检测,有效提高了语音段起止端点检测的精确度和可靠性。实验结果表明,该方法比短时能量方法的计算效率高,在信噪比为-10 dB时仍能完成端点检测。
An algorithm that combines the expanded spectral subtraction with the SAP （Speech Absence Probability） soft decision is proposed to improve the capability of endpoint detection for lower RSNR. It uses a method of expanded spectral subtraction based on the noise compensation structure ruled by an adaptive decision, which eliminates the noise efficiently without voice activity detector, and also overcomes the typical disadvantage of one channel noise suppression algorithm, which can not estimate the noise during speech presence. At the same time, a method of end point detection based on the SAP soft decision is given, which improves the robustness and the precision of the endpoint detection. Compared to the method based on energy, the experiments show that higher efficiency and good detection capability can be obtained under the condition of RSNR equal to - 10 dB.
出处
《吉林大学学报（信息科学版）》
CAS
2006年第4期351-357,共7页
Journal of Jilin University（Information Science Edition）
基金
长春市科技计划基金资助项目(05GG18)
关键词
扩展谱相减法
噪声消除
SAP软门限
端点检测
expanded spectral subtraction
noise reduction
speech absence probability （SAP） soft decision
endpoint detection
