NOISE SUPPRESSION DEVICE AND NOISE SUPPRESSION METHOD

• US 20100262425A1
• Filed: 03/18/2009
• Published: 10/14/2010
• Est. Priority Date: 03/21/2008
• Status: Active Grant

• Current Assignee: Tokyo University of Science Educational Foundation Administrative Organization
• Original Assignee: Tokyo University of Science Educational Foundation Administrative Organization
• Inventors: Tanabe, Nari, Furukawa, Toshihiro
• Granted Patent: US 8,527,266 B2
• Time in Patent Office: Days
• US Class Current: 704/233
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 2207/20076 Probabilistic image processing
• CPC Class Codes: G06T 2207/20182 Noise reduction or smoothing in the temporal domain; Spatio-temporal filtering
• CPC Class Codes: G06T 2207/20221 Image fusion; Image merging
• CPC Class Codes: G06T 5/50 using two or more images, e.g. averaging or subtraction
• CPC Class Codes: G06T 5/70 Denoising; Smoothing
• CPC Class Codes: G06T 5/73 Deblurring; Sharpening
• CPC Class Codes: G10L 2021/02163 Only one microphone
• CPC Class Codes: G10L 21/0208 Noise filtering
• CPC Class Codes: G10L 25/06 the extracted parameters being correlation coefficients

Abstract

Disclosed is a noise suppression device capable of better noise suppression by means of a simpler structure and with a lighter computational load. A noise suppression device (100) has a noise suppression processor (150) to estimate the required information only from the observed information, which is the required information corrupted by noise. A correlator (154) calculates the correlation of the estimation error when the state quantity, which contains the required information, of the system at time n+1 was estimated from the information until time n or time n+1 for the observed information at only time n. A weighting coefficient calculator (156) uses the correlation calculated by the correlator (154) for the observed information at only time n to calculate the weighting coefficients for specifying the relationships of the optimum estimate of the state quantity at that time based on the information until time n+1, the optimum estimate of the state quantity at time n+1 based on the information until time n, and the estimation error of the observed quantity including the observed information. An optimum estimate calculator (158) uses the weighting coefficients calculated by the weighting coefficient calculator (156) for the observed information only at time n to calculate the optimum estimate of the state quantity at that time based on the information until time n or time n+1.

Claims

1. A noise suppression apparatus that estimates desired information from only observed information in which noise is mixed in with the desired information, the noise suppression apparatus comprising:

a correlation computation section that calculates a correlation value of estimation error when a system state quantity of time n+1 that includes the desired information is estimated based on information up to time n or time n+1 for observed information of only time n;

a weighting coefficient calculation section that uses a correlation value calculated by the correlation computation section for observed information of only time n to calculate a weighting coefficient for specifying relationships of an optimum estimate of the state quantity at that time based on information up to time n+1, an optimum estimate of the state quantity at time n+1 based on information up to time n, and estimation error of an observed quantity including the observed information; and

an optimum estimate calculation section that uses a weighting coefficient calculated by the weighting coefficient calculation section for observed information of only time n to calculate an optimum estimate of the state quantity at that time based on information up to time n or time n+1.

2. The noise suppression apparatus according to claim 1, wherein:

the correlation computation section calculates as a scalar quantity a correlation value of estimation error when a system state quantity of time n+1 that includes the desired information is estimated based on information up to time n for observed information of only time n;

the weighting coefficient calculation section uses the scalar quantity of the correlation value of estimation error calculated by the correlation computation section for observed information of only time n to calculate as a scalar quantity a weighting coefficient for specifying relationships of an optimum estimate of the state quantity at that time based on information up to time n+1, an optimum estimate of the state quantity at time n+1 based on information up to time n, and estimation error of an observed quantity including the observed information; and

the optimum estimate calculation section uses the scalar quantity of the weighting coefficient calculated by the weighting coefficient calculation section for observed information of only time n to calculate as a scalar quantity an optimum estimate of the state quantity at that time based on information up to time n+1.

3. The noise suppression apparatus according to claim 2, wherein:

the correlation computation section uses a provided driving source covariance scalar quantity to calculate a scalar quantity of a correlation value of the estimation error;

the weighting coefficient calculation section uses a provided covariance scalar quantity of noise and the scalar quantity of the correlation value of estimation error calculated by the correlation computation section to calculate a scalar quantity of the weighting coefficient; and

the optimum estimate calculation section uses the scalar quantity of the weighting coefficient calculated by the weighting coefficient calculation section and an observed quantity at only time n+1 to calculate a scalar quantity of an optimum estimate of the state quantity.

4. The noise suppression apparatus according to claim 1, wherein:

the correlation computation section has a first correlation computation section and a second correlation computation section;

the optimum estimate calculation section has a first optimum estimate calculation section and a second optimum estimate calculation section;

the first correlation computation section calculates as a matrix a correlation value of estimation error when a system state quantity of time n+1 that includes the desired information is estimated based on information up to time n for observed information of only time n;

the weighting coefficient calculation section uses the matrix of the correlation value of estimation error calculated by the first correlation computation section for observed information of only time n to calculate as a vector quantity a weighting coefficient for specifying relationships of an optimum estimate of the state quantity at that time based on information up to time n+1, an optimum estimate of the state quantity at time n+1 based on information up to time n, and estimation error of an observed quantity including the observed information;

the first optimum estimate calculation section calculates as a vector quantity an optimum estimate of the state quantity at time n+1 based on information up to time n for observed information of only time n;

the second optimum estimate calculation section uses the vector quantity of the weighting coefficient calculated by the weighting coefficient calculation section for observed information of only time n to calculate as a vector quantity an optimum estimate of the state quantity at that time based on information up to time n+1; and

the second correlation computation section calculates as a matrix a correlation value of estimation error when a state quantity of that time is estimated based on information up to time n+1 for observed information of only time n.

5. The noise suppression apparatus according to claim 4, wherein:

the first correlation computation section uses a predetermined state transition matrix, a provided covariance element of a driving source vector, and the matrix of the correlation value of estimation error that has been provided or that was calculated last time by the second correlation computation section to calculate the matrix of the correlation value of estimation error;

the weighting coefficient calculation section uses the matrix of the correlation value of estimation error calculated by the first correlation computation section, a provided observation transition vector, and a provided covariance scalar quantity of noise to calculate a vector quantity of the weighting coefficient;

the first optimum estimate calculation section uses the state transition matrix and the vector quantity of the optimum estimate of the state quantity that has been provided or was calculated last time by the second optimum estimate calculation section to calculate the vector quantity of the optimum estimate of the state quantity;

the second optimum estimate calculation section uses the vector quantity of the optimum estimate of the state quantity calculated by the first optimum estimate calculation section, the vector quantity of the weighting coefficient calculated by the weighting coefficient calculation section, the observation transition vector, and an observed quantity at only time n+1 to calculate the vector quantity of the optimum estimate of the state quantity; and

the second correlation computation section uses the vector quantity of the weighting coefficient calculated by the weighting coefficient calculation section, the observation transition vector, and the matrix of the correlation value of estimation error calculated by the first correlation computation section to calculate the matrix of the correlation value of estimation error.

12. A fetal heart sound detection apparatus comprising:

the noise suppression apparatus according to claim 1 that takes a signal obtained by detecting a fetal heartbeat as observed information, and estimates a fetal heartbeat as desired information; and

an output section that outputs a fetal heartbeat estimated by the noise suppression apparatus.

13. A portable terminal apparatus comprising:

a microphone that detects voice of a user as an electrical speech signal;

a receiving section that receives a speech signal of a communicating party transmitted as a radio signal;

the noise suppression apparatus according to claim 1 that takes the speech signal detected by the microphone or the speech signal received by the receiving section as observed information, and estimates the speech signal of the user or the communicating party as desired information;

a transmitting section that transmits as a radio signal the speech signal of the user estimated by the noise suppression apparatus; and

a speaker that outputs the speech signal of the communicating party estimated by the noise suppression apparatus.

14. A car navigation apparatus comprising:

a microphone that detects voice of a user as an electrical speech signal;

the noise suppression apparatus according to claim 1 that takes the speech signal detected by the microphone as observed information, and estimates the speech signal of the user as desired information;

a speech recognition section that has the speech signal of the user estimated by the noise suppression apparatus as input, and performs speech recognition processing; and

a car navigation processing section that takes a recognition result of the speech recognition section as an operation command, and executes car navigation processing.

15. A speech recognition apparatus comprising:

the noise suppression apparatus according to claim 1 that takes a speech signal obtained by detecting voice of a user as observed information, and estimates the speech signal of the user as desired information;

a speech recognition section that has the speech signal of the user estimated by the noise suppression apparatus as input, and performs speech recognition processing; and

an output section that outputs a recognition result of the speech recognition section.

16. An abnormality detection apparatus comprising:

the noise suppression apparatus according to claim 1 that takes a signal obtained by detecting sound emitted by a test object as observed information, and estimates test object information as desired information;

an abnormal sound analysis section that detects abnormal sound by comparing desired information estimated by the noise suppression apparatus with normal sound; and

an output section that outputs an analysis result of the abnormal sound analysis section.

17. An image processing apparatus comprising:

an image input section that has an image signal as input;

the noise suppression apparatus according to claim 1 that takes an image signal input by means of the image input section as observed information, and estimates an original image as desired information; and

an output section that outputs an original image estimated by the noise suppression apparatus.

6. A noise suppression method that estimates desired information from only observed information in which noise is mixed in with the desired information, the noise suppression method comprising:

a correlation computation step of calculating a correlation value of estimation error when a system state quantity of time n+1 that includes the desired information is estimated based on information up to time n or time n+1 for observed information of only time n;

a weighting coefficient calculation step of using a correlation value calculated by the correlation computation step for observed information of only time n to calculate a weighting coefficient for specifying relationships of an optimum estimate of the state quantity at that time based on information up to time n+1, an optimum estimate of the state quantity at time n+1 based on information up to time n, and estimation error of an observed quantity including the observed information; and

an optimum estimate calculation step of using a weighting coefficient calculated by the weighting coefficient calculation step for observed information of only time n to calculate an optimum estimate of the state quantity at that time based on information up to time n or time n+1.

7. The noise suppression method according to claim 6, wherein:

the correlation computation step calculates as a scalar quantity a correlation value of estimation error when a system state quantity of time n+1 that includes the desired information is estimated based on information up to time n for observed information of only time n;

the weighting coefficient calculation step uses the scalar quantity of the correlation value of estimation error calculated by the correlation computation step for observed information of only time n to calculate as a scalar quantity a weighting coefficient for specifying relationships of an optimum estimate of the state quantity at that time based on information up to time n+1, an optimum estimate of the state quantity at time n+1 based on information up to time n, and estimation error of an observed quantity including the observed information; and

the optimum estimate calculation step uses the scalar quantity of the weighting coefficient calculated by the weighting coefficient calculation step for observed information of only time n to calculate as a scalar quantity an optimum estimate of the state quantity at that time based on information up to time n+1.

8. The noise suppression method according to claim 6, wherein:

the correlation computation step has a first correlation computation step and a second correlation computation step;

the optimum estimate calculation step has a first optimum estimate calculation step and a second optimum estimate calculation step;

the first correlation computation step calculates as a matrix a correlation value of estimation error when a system state quantity of time n+1 that includes the desired information is estimated based on information up to time n for observed information of only time n;

the weighting coefficient calculation step uses the matrix of the correlation value of estimation error calculated by the first correlation computation step for observed information of only time n to calculate as a vector quantity a weighting coefficient for specifying relationships of an optimum estimate of the state quantity at that time based on information up to time n+1, an optimum estimate of the state quantity at time n+1 based on information up to time n, and estimation error of an observed quantity including the observed information;

the first optimum estimate calculation step calculates as a vector quantity an optimum estimate of the state quantity at time n+1 based on information up to time n for observed information of only time n;

the second optimum estimate calculation step uses the vector quantity of the weighting coefficient calculated by the weighting coefficient calculation step for observed information of only time n to calculate as a vector quantity an optimum estimate of the state quantity at that time based on information up to time n+1; and

the second correlation computation step calculates as a matrix a correlation value of estimation error when a state quantity of that time is estimated based on information up to time n+1 for observed information of only time n.

9. A noise suppression program for estimating desired information from only observed information in which noise is mixed in with the desired information, the noise suppression program causing a computer to execute:

a correlation computation step of calculating a correlation value of estimation error when a system state quantity of time n+1 that includes the desired information is estimated based on information up to time n or time n+1 for observed information of only time n;

a weighting coefficient calculation step of using a correlation value calculated by the correlation computation step for observed information of only time n to calculate a weighting coefficient for specifying relationships of an optimum estimate of the state quantity at that time based on information up to time n+1, an optimum estimate of the state quantity at time n+1 based on information up to time n, and estimation error of an observed quantity including the observed information; and

an optimum estimate calculation step of using a weighting coefficient calculated by the weighting coefficient calculation step for observed information of only time n to calculate an optimum estimate of the state quantity at that time based on information up to time n or time n+1.

10. The noise suppression program according to claim 9, wherein:

the correlation computation step calculates as a scalar quantity a correlation value of estimation error when a system state quantity of time n+1 that includes the desired information is estimated based on information up to time n for observed information of only time n;

the weighting coefficient calculation step uses the scalar quantity of the correlation value of estimation error calculated by the correlation computation step for observed information of only time n to calculate as a scalar quantity a weighting coefficient for specifying relationships of an optimum estimate of the state quantity at that time based on information up to time n+1, an optimum estimate of the state quantity at time n+1 based on information up to time n, and estimation error of an observed quantity including the observed information; and

the optimum estimate calculation step uses the scalar quantity of the weighting coefficient calculated by the weighting coefficient calculation step for observed information of only time n to calculate as a scalar quantity an optimum estimate of the state quantity at that time based on information up to time n+1.

11. The noise suppression program according to claim 9, wherein:

the correlation computation step has a first correlation computation step and a second correlation computation step;

the optimum estimate calculation step has a first optimum estimate calculation step and a second optimum estimate calculation step;

the first correlation computation step calculates as a matrix a correlation value of estimation error when a system state quantity of time n+1 that includes the desired information is estimated based on information up to time n for observed information of only time n;

the weighting coefficient calculation step uses the matrix of the correlation value of estimation error calculated by the first correlation computation step for observed information of only time n to calculate as a vector quantity a weighting coefficient for specifying relationships of an optimum estimate of the state quantity at that time based on information up to time n+1, an optimum estimate of the state quantity at time n+1 based on information up to time n, and estimation error of an observed quantity including the observed information;

the first optimum estimate calculation step calculates as a vector quantity an optimum estimate of the state quantity at time n+1 based on information up to time n for observed information of only time n;

the second optimum estimate calculation step uses the vector quantity of the weighting coefficient calculated by the weighting coefficient calculation step for observed information of only time n to calculate as a vector quantity an optimum estimate of the state quantity at that time based on information up to time n+1; and

the second correlation computation step calculates as a matrix a correlation value of estimation error when a state quantity of that time is estimated based on information up to time n+1 for observed information of only time n.