摘要 |
PROBLEM TO BE SOLVED: To provide the voice signal analyzer in which the constitution is made simple compared with conventional technology and the polarities of voice signals are discriminated with high measurement precision. SOLUTION: Signal waveforms s(n) of the voice digital signals, which are obtained by A/D-transforming voice signals, are analyzed in a linear prediction analysis section 4 by an LPC analysis method and the coefficients of the vocal tract transfer function, which represents the vocal tract characteristics, are obtained. In a sound source waveform extracting section 5, inverse filter waveforms v(n), which include noise components, are generated using an inverse filter method, the low frequency noise components are eliminated through a low-pass filter(LPF) 6 and pseudovocal chink sound source waveforms SG(n) are obtained. In a gradient computing section 8 and an absolute value computing section 9, the waveforms SG(n) are differentiated and absolute values are computed and absolute value signals AGSG(n) of the gradients of the inverse filter waveforms are obtained. A polarity discriminating section 10 determines the polarities of voices by employing the waveforms SG(n) and the signals AGSG(n) of the gradients and outputs the signals which indicate the results. |