Back to results
US 9,318,120 B2Granted
System and method for noise reduction in processing speech signals by targeting speech and disregarding noise
AssigneeSCTI Holdings, Inc.
InventorMark Pinson
Life of this patent
Estimating the term…
Abstract
An exemplary noise reduction system and method processes a speech signal that is delivered in a noisy channel or with ambient noise. Some exemplary embodiments of the system and method use filters to extract speech information, and focus on a subset of harmonics that are least corrupted by noise. Some exemplary embodiments disregard signal harmonics with low signal-to-noise ratio(s), and disregard amplitude modulations that are inconsistent with speech. An exemplary system and method processes a signal that focuses on a subset of harmonics that are least corrupted by noise, disregards the signal harmonics with low signal-to-noise ratio(s), and disregards amplitude modulations that are inconsistent with speech.
Claims24
A method of targeting speech and disregarding noise, comprising:
- receiving an input signal, wherein the input signal includes a human speech signal, and noise other than the human speech signal, wherein the input signal has a fundamental frequency;
- identifying the fundamental frequency of the input signal;
- identifying one or more harmonics of the fundamental frequency;
- applying a filter to the fundamental frequency and the one or more harmonics, thereby attenuating energy other than harmonic energy and forming one or more harmonic-only signals;
- performing an amplitude modulation on the one or more harmonic-only signals, thereby forming one or more amplitude modulated harmonic-only signals;
- constructing one or more signal-to-noise ratios for the one or more amplitude modulated harmonic-only signals;
- disregarding one or more signals from among the one or more amplitude modulated harmonic-only signals whose signal-to-noise ratio falls below a threshold amplitude, thereby forming extracted amplitude patterns;
- applying a bandpass filter to the extracted amplitude patterns, the bandpass filter having a range characteristic of human speech, thereby disregarding one or more signals from among the one or more amplitude modulated harmonic-only signals whose amplitude modulations are inconsistent with the human speech;
- and
- outputting a resultant signal.
7 more elements
An apparatus for automatic speech extraction comprising:
- an input device configured for receiving an input signal, wherein the input signal includes a human speech signal, and noise other than the human speech signal;
- and
- a processor operatively coupled with a memory device and with the input device, wherein the processor includes a plurality of processing modules comprising:
- a harmonic frequency identifier configured for identifying the fundamental frequency of the input signal;
- a filter applicator configured for applying a filter to the fundamental frequency and the one or more harmonics, thereby attenuating energy other than harmonic energy and forming one or more harmonic-only signals;
- a modulator configured for performing an amplitude modulation on the one or more harmonic-only signals to form one or more amplitude modulated harmonic-only signals;
- a power ratio constructor configured for constructing one or more signal to noise ratios for the one or more amplitude modulated harmonic-only signals;
- a low power harmonic remover configured for disregarding one or more signals from among the one or more amplitude modulated harmonic-only signals whose signal to noise ratio falls below a threshold amplitude, to form extracted amplitude patterns;
- and
- a non-speech harmonic remover comprising a bandpass filter having a range characteristic of human speech for application to the extracted amplitude patterns, wherein the non-speech harmonic remover is configured to disregard one or more signals from among the one or more amplitude modulated harmonic-only signals whose amplitude modulations are inconsistent with the human speech;
- and
- a processed signal output.
9 more elements
A method, comprising:
- receiving an input signal, wherein the input signal includes a human speech signal, and noise other than the human speech signal, wherein the input signal has a fundamental frequency;
- identifying the fundamental frequency of the input signal;
- identifying one or more harmonics of the fundamental frequency;
- applying a filter to the fundamental frequency and the one or more harmonics, thereby attenuating energy other than harmonic energy and forming one or more harmonic-only signals;
- performing an amplitude modulation on the one or more harmonic-only signals, thereby forming one or more amplitude modulated harmonic-only signals;
- constructing one or more signal-to-noise ratios for the one or more amplitude modulated harmonic-only signals;
- disregarding one or more signals from among the one or more amplitude modulated harmonic-only signals whose signal-to-noise ratio falls below a threshold amplitude, and further comprising:
- defining at least one wide band centered on the fundamental frequency and extending from substantially halfway to the next lower harmonic frequency and to substantially halfway to the next higher harmonic frequency;
- defining at least one narrow band centered on the fundamental frequency, wherein the narrow band's bandwidth is less than the wide band's bandwidth;
- computing a ratio of energy in the narrow band to energy in the wide band;
- disregarding one or more signals from among the one or more amplitude modulated harmonic-only signals whose ratio of energy in the narrow band to energy in the wide band is higher than a threshold value;
- estimating the frequency of one or more harmonics based on the fundamental frequency;
- determining the actual frequency of the one or more harmonics observed in the identifying one or more harmonics of the fundamental frequency;
- determining a deviation value between the estimated frequency of the one or more harmonics and the actual frequency of the one or more harmonics;
- and
- disregarding one or more signals from among the one or more amplitude modulated harmonic-only signals whose deviation value exceeds a predetermined value, thereby forming extracted amplitude patterns;
- and
- applying a bandpass filter to the extracted amplitude patterns, the bandpass filter having a range characteristic of human speech, thereby disregarding one or more signals from among the one or more amplitude modulated harmonic-only signals whose amplitude modulations are inconsistent with the human speech;
- wherein a resultant signal is output.
16 more elements
Dependencies are inferred from the claim text.