A VOICE ACTIVITY DETECTION METHOD BASED ON NON-NEGATIVE MATRIX FACTORIZATION

摘要

PURPOSE: A voice detecting method based on a non-negative matrix factorization is provided to estimate noise according to error value distribution of a noise estimation section to apply an optimum threshold value to an error, thereby detecting a voice signal. CONSTITUTION: A basic vector is made by a noise signal and an input signal using NMF(Non-negative Matrix Factorization)(S100). The error between a noise signal basic vector and an input signal basic vector is calculated(S200). An optimized threshold value is obtained(S300). A voice is detected by applying the threshold value to the calculated error(S400).