| 摘要 |
PURPOSE:To surely detect sounds independently of the input level or the background noise by providing a feature parameter calculator, a buffer, a threshold comparator, and a threshold generator. CONSTITUTION:The input signal from a terminal 100 is sampled with 8kHz, and every 160 samples are gathered as one frame, and a linear forecast coefficient is calculated with one frame as the unit by a feature parameter calculator 101 to obtain feature parameters X(n). Feature parameters X(n) are stored in a buffer 109 in the input order. In a threshold generator 110, a normalization coefficient calculator calculates an average value and a standard deviation of a set of feature parameters corresponding to N frames, and a threshold calculator obtains a threshold T(n). A threshold comparator 108 discriminates noise in the case of X(n)<T(n) and discriminates sounds in the case of X(n)>=T(n). Since this constitution emphasizes features of sounds and noise based on feature parameters in the buffer, the sound recognition rate is improved and the tone quality is improved. |
