Voice feature extraction device

A voice feature extraction device comprising:

• a noise reduction system coefficient calculation unit that adds a gain-adjusted simulated voice signal to a surrounding signal, and calculates a noise reduction system coefficient of a noise reduction system based on the added signal and the gain-adjusted simulated voice signal delayed by a delay processing unit, and
• a microphone that collects an input voice signal of a user,
• a window function operation unit that samples the voice signal inputted from the microphone, and prevents generation of high frequency components caused by a data jump at intervals of each frame,
• an input voice signal power spectrum calculation unit that calculates a power spectrum vector of the input voice signal processed by the window function operation unit, and
• a noise reduction system that is set to the coefficient calculated by the noise reduction system coefficient calculation unit, and executes a noise reduction processing on the power spectrum vector.

A method of extracting voice features comprising:

• adding a gain-adjusted simulated voice signal to a surrounding signal;
• calculating a noise reduction system coefficient of a noise reduction system to be used based on the added signal and the gain-adjusted simulated voice signal delayed by a delay processing unit, and
• calculating a power spectrum vector of a power spectrum signal produced from an input voice signal,
• wherein the noise reduction system that is set to the calculated noise reduction system coefficient executes a noise reduction processing on the power spectrum vector, and extracts the voice features.

A method of extracting voice features comprising:

• adding a gain-adjusted simulated voice signal to a surrounding signal;
• calculating a noise reduction system coefficient of a noise reduction system based on the added signal and the gain-adjusted simulated voice signal delayed by a delay processing unit, and
• sampling a voice signal inputted from a microphone,
• executing a processing to prevent generation of high frequency components of the input voice signal sampled,
• calculating a power spectrum vector of a power spectrum signal produced from the input voice signal that is processed to prevent generation of high frequency components, and
• calculating a voice feature from the power spectrum vector via the noise reduction system that is set to the calculated noise reduction system coefficient.

A voice feature extraction device comprising:

• a microphone that collects a surrounding signal;
• a simulated voice signal generation unit that generates a simulated voice signal;
• a gain adjustment unit that adjusts a gain of the simulated voice signal;
• an adder that adds the surrounding signal collected by the microphone and the gain-adjusted simulated voice signal;
• a delay processing unit that delays the gain-adjusted simulated voice signal by a predetermined time;
• an adaptive filter that executes an adaptive control on the signals added by the adder and the simulated voice signal delayed by the delay processing unit, and generates a filter coefficient;
• an FFT operation unit that executes a fast Fourier transform on the filter coefficient generated by the adaptive control of the adaptive filter;
• a power calculation unit that calculates a power spectrum vector from a power spectrum signal calculated by the FFT operation unit;
• and
• a noise reduction system having the power spectrum vector calculated by the power calculation unit set as a noise reduction coefficient.