# US 9,881,633 B2 · Audio signal processing device, audio signal processing method, and audio signal processing program

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 9,881,633 B2
- Document ID: US9881633B2
- Status: Granted
- Kind: B2
- Application number: 15/503,297
- Filed: 2014-09-12
- Published: 2017-08-17
- Granted: 2018-01-30
- Earliest claimed: 2014-08-14
- Assignees: P Softhouse Co., Ltd.
- Applicants: P Softhouse Co., Ltd.
- Inventors: Takuma Kudou
- Web page: https://patentgrep.com/patent/US9881633B2

## CPC

- G10L 21/0272: Voice signal separating
- G10L 21/0232: Processing in the frequency domain
- G10L 21/0208: Noise filtering
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

An audio signal processing device includes: a short-time fast Fourier transform unit that generates a signal in a frequency domain obtained by performing a short-time fast Fourier transform on an input audio signal; a steady sound determining unit that determines whether a waveform of a peak portion included in a waveform of the signal in a frequency domain is a steady sound; a filter coefficient calculation unit that dynamically calculates a filter coefficient on the basis of a result of determination made by the steady sound determining unit; a comb filter that operates according to the filter coefficient calculated by the filter coefficient calculation unit so as to filter a signal in a frequency domain; and an inverse Fourier transform unit that transforms an output of the comb filter into a signal in a time domain and outputs the signal in a time domain.

## Claims

### Claim 1

An audio signal processing device that separates a specific sound source from an audio signal in which a plurality of sound sources are mixed and extracts or eliminates the specific sound source, the audio signal processing device comprising:
a short-time fast Fourier transform unit that performs a short-time fast Fourier transform on an input audio signal;
a steady sound determining unit that includes a smoothing processing unit that applies a low pass filter to a signal in a frequency domain generated by the short time fast Fourier transform unit to smooth the signal in a frequency domain and a peak sharpness determining unit that determines a sharpness of a waveform of a peak portion included in a waveform of the signal in a frequency domain on a basis of an output difference between the signal in a frequency domain and a signal output from the smoothing processing unit and that determines whether the waveform of the peak portion included in the waveform of the signal in a frequency domain is a steady sound;
a filter coefficient calculation unit that dynamically calculates a filter coefficient on a basis of a result of determination made by the steady sound determining unit;
a comb filter that operates according to the filter coefficient calculated by the filter coefficient calculation unit so as to filter a signal output from the short-time fast Fourier transform unit; and
an inverse Fourier transform unit that transforms an output of the comb filter into a signal in a time domain and outputs the signal in a time domain, wherein
when the low pass filter is applied, the steady sound determining unit adjusts the filter coefficient such that the higher a frequency band is, the smoother the waveform of the signal is.

### Claim 2 (depends on 1)

The audio signal processing device according to claim 1, wherein the filter coefficient of the comb filter is dynamically constructed according to a filter coefficient of the low pass filter.

### Claim 3

An audio signal processing method of separating a specific sound source from an audio signal in which a plurality of sound sources are mixed and extracting or eliminating the specific sound source, the audio signal processing method comprising:
a first step of performing a short-time fast Fourier transform on an input audio signal;
a second step of applying a low pass filter to a signal in a frequency domain generated at the first step to smooth the signal in a frequency domain;
a third step of determining a sharpness of a waveform of a peak portion included in a waveform of the signal in a frequency domain on a basis of an output difference between the signal in a frequency domain and a signal output at the second step;
a fourth step of determining whether the waveform of the peak portion is a steady sound on a basis of a result of determination at the third step;
a fifth step of dynamically calculating a filter coefficient for comb filtering on a basis of a result of determination at the fourth step;
a sixth step of filtering the signal in a frequency domain generated at the first step using the filter coefficient calculated at the fifth step; and
a seventh step of transforming an output of filtering at the sixth step into a signal in a time domain and outputting the signal in a time domain, wherein
the second step includes, when applying the low pass filter, adjusting the filter coefficient such that the higher a frequency band is, the smoother the waveform of the signal is.

### Claim 4 (depends on 3)

The audio signal processing method according to claim 3, wherein the filter coefficient for comb filtering is dynamically determined according to a filter coefficient of the low pass filter.

### Claim 5

An audio signal processing method of separating a specific sound source from an audio signal in which a plurality of sound sources are mixed and extracting or eliminating the specific sound source, the audio signal processing method comprising:
a first step of performing a short-time fast Fourier transform on an input audio signal;
a second step of evaluating, for a waveform of a peak portion included in a waveform of a signal in a frequency domain, an amplitude drop rate that is a ratio of a drop amount from a peak value of the peak portion in a preset frequency width to the frequency width;
a third step of determining, on a basis of a result of evaluation at the second step, whether the waveform of the peak portion is a steady sound;
a fourth step of dynamically calculating a filter coefficient for comb filtering on a basis of a result of determination at the third step;
a fifth step of filtering the signal in a frequency domain generated at the first step using the filter coefficient calculated at the fourth step; and
a sixth step of transforming an output of filtering at the fifth step into a signal in a time domain and outputting the signal in a time domain, wherein
the second step includes, when evaluating the amplitude drop rate, adjusting the filter coefficient such that the higher a frequency band is, the smaller an evaluated value of the amplitude drop.

### Claim 6 (depends on 5)

A non-transitory computer-readable recording medium that stores therein an audio signal processing program that causes a processor to execute the audio signal processing method according to claim 5.