WO2021235750A1 - 강인음성인식을 위한 방향벡터 추정을 겸한 온라인 우도최대화를 이용한 빔포밍 방법 및 그 장치 - Google Patents
강인음성인식을 위한 방향벡터 추정을 겸한 온라인 우도최대화를 이용한 빔포밍 방법 및 그 장치 Download PDFInfo
- Publication number
- WO2021235750A1 WO2021235750A1 PCT/KR2021/005759 KR2021005759W WO2021235750A1 WO 2021235750 A1 WO2021235750 A1 WO 2021235750A1 KR 2021005759 W KR2021005759 W KR 2021005759W WO 2021235750 A1 WO2021235750 A1 WO 2021235750A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- current frame
- covariance
- beamforming
- results
- variance
- Prior art date
Links
- 239000013598 vector Substances 0.000 title claims abstract description 224
- 238000000034 method Methods 0.000 title claims description 12
- 238000000605 extraction Methods 0.000 claims abstract description 96
- 238000010586 diagram Methods 0.000 description 48
- BQCADISMDOOEFD-UHFFFAOYSA-N Silver Chemical compound [Ag] BQCADISMDOOEFD-UHFFFAOYSA-N 0.000 description 6
- 229910052709 silver Inorganic materials 0.000 description 6
- 239000004332 silver Substances 0.000 description 6
- 230000001419 dependent effect Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000009792 diffusion process Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 239000006185 dispersion Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L2021/02082—Noise filtering the noise being echo, reverberation of the speech
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
- G10L2021/02161—Number of inputs available containing the signal or the noise to be suppressed
- G10L2021/02166—Microphone arrays; Beamforming
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0272—Voice signal separating
- G10L21/0308—Voice signal separating characterised by the type of parameter measurement, e.g. correlation techniques, zero crossing techniques or predictive techniques
Definitions
- the present invention relates to a beamforming method and apparatus using online likelihood maximization with direction vector estimation for robust speech recognition.
- the sound input signal input through the microphone may include not only the target voice required for voice recognition but also noises that interfere with voice recognition.
- Various studies are being conducted to improve the performance of voice recognition by removing noise from the sound input signal and extracting only the desired target voice.
- the technical problem to be achieved by the present invention is to generate a direction vector by calculating the noise covariance based on the variance determined according to the output results corresponding to the input results, and to improve the extraction performance for the target sound source by updating the beamforming weight. It is to provide a target signal extraction device that can.
- the apparatus for extracting a target signal may include a direction vector predictor and a beamformer.
- the direction vector predictor generates an input signal covariance according to input results for each frequency over time, and generates a noise covariance based on a variance determined according to output results corresponding to the input results, the input signal covariance and A direction vector may be generated based on the noise covariance.
- the beamformer may generate a beamforming weight according to the beamforming covariance and the direction vector determined according to the variance, and provide the output results based on the input results and the beamforming weight.
- the variance of the noise covariance and the beamforming covariance may be determined based on output results.
- the initial values of the noise covariance and the beamforming covariance may be determined based on the input results.
- the noise covariance may be determined according to a larger value among the variance and the first constant value.
- the noise covariance may be normalized according to a larger value among the variance and the first constant value.
- the beamforming covariance may be determined according to a larger value of the variance and a second constant value.
- the target signal extraction apparatus may repeatedly operate the direction vector predictor and the beamformer until the beamforming weights converge.
- the target signal extraction system may include a direction vector predictor and a beamformer.
- the direction vector predictor generates an input signal covariance according to input results for each frequency over time, and generates a noise covariance based on a variance determined according to output results corresponding to the input results and a predetermined mask, and the A direction vector may be generated based on the input signal covariance and the noise covariance.
- the beamformer may generate a beamforming weight according to the beamforming covariance and the direction vector determined according to the variance, and provide the output results based on the input results and the beamforming weight.
- the initial values of the noise covariance and the beamforming covariance may be determined according to a product of the input results and the mask.
- the input results of the noise covariance may be updated as a product of the input results and the mask.
- the mask may be calculated for each frame index and frequency index.
- the noise covariance may be determined according to a larger value of the variance and a first constant value, and the noise covariance may be normalized according to a larger value of the variance and the first constant value. have.
- the beamforming covariance is determined according to a larger value of the variance and a second constant value, and the target signal extracting apparatus uses the direction vector predictor and the beamformer until the beamforming weights converge. It can be operated repeatedly.
- the online target signal extraction apparatus may include a direction vector predictor and a beamformer.
- the direction vector predictor generates the current frame input signal covariance generated based on the previous frame input signal covariance corresponding to the previous frame and the current frame input results for each frequency according to the current frame, and the previous frame noise covariance corresponding to the previous frame , generates a current frame noise covariance based on a current frame variance estimation value generated according to current frame input results corresponding to the current frame and a previous frame beamforming weight corresponding to the previous frame, the current frame input signal covariance, the current frame
- the current frame direction vector may be generated based on the frame noise covariance and the previous frame direction vector corresponding to the previous frame.
- the beamformer generates a current frame beamforming variance estimation value generated according to the previous frame beamforming weight corresponding to the previous frame, the previous frame variance corresponding to the current frame input results and the previous frame output results, and corresponds to the previous frame.
- a current frame beamforming inverse covariance is generated according to the previous frame inverse covariance, the current frame input results, and the current frame beamforming variance estimation value, and the current frame beam according to the current frame direction vector and the current frame inverse covariance
- a forming weight may be generated, and the current frame output results may be provided based on the current frame input results and the current frame beamforming weight.
- the current frame noise covariance may be normalized by a current frame variance estimate.
- the online target signal extraction system may include a direction vector predictor and a beamformer.
- the direction vector predictor generates the current frame input signal covariance generated based on the previous frame input signal covariance corresponding to the previous frame and the current frame input results for each frequency according to the current frame, and the previous frame noise covariance corresponding to the previous frame , the current frame input results corresponding to the current frame and generating a current frame noise covariance through a current frame variance estimate generated according to a predetermined mask, and a current frame direction based on the current frame input signal covariance, the current frame noise covariance, and a previous frame direction vector corresponding to the previous frame.
- the beamformer generates a beamforming variance estimate of the current frame through the previous frame beamforming weight corresponding to the previous frame, the current frame input results, the previous frame variance corresponding to the previous frame output results, and a predetermined mask, and A current frame beamforming inverse covariance is generated according to the corresponding previous frame inverse covariance, the current frame input results, and the current frame beamforming variance estimation value, and the current frame beam according to the current frame direction vector and the current frame beamforming inverse covariance.
- a forming weight may be generated, and the current frame output results may be provided based on the current frame input results and the current frame beamforming weight.
- the current frame noise covariance may be generated based on the previous frame noise covariance and the current frame input results and a current frame variance estimation value generated through a predetermined mask.
- the current frame beamforming variance estimation value may be generated based on the previous frame beamforming weight, the current frame input results, the previous frame variance, and a predetermined mask.
- the weighted covariance and the weighted correlation vector may be determined according to a larger value of a variance and a second constant value, and the target signal extraction system may converge the de-echo filter and the beamforming weight.
- the echo canceller, the direction vector predictor, and the beamformer may be repeatedly operated until the
- the apparatus for extracting a target signal may include an echo canceller, an echo vector predictor, and a beamformer.
- the echo canceller generates a weighted covariance based on a variance determined according to the past input results for each frequency over time and output results corresponding to the echo-cancelled input results, and generates the weighted covariance with the input results for each frequency over time and the past.
- the direction vector predictor generates an input signal covariance according to the de-echoed input results, and generates a noise covariance based on a variance determined according to output results corresponding to the input results, the input signal covariance and the A direction vector can be generated based on the noise covariance.
- the beamformer may generate a beamforming weight according to the beamforming covariance and the direction vector determined according to the variance, and provide the output results based on the echo-cancelled input results and the beamforming weight.
- the weighted covariance, the weighted correlation vector, the noise covariance, and the beamforming covariance may be determined based on the output results.
- the weighted covariance and initial values of the weighted correlation vector may be determined based on the input results.
- the weighted covariance and the weighted correlation vector may be determined according to a larger value of the variance and the second constant value.
- the initial values of the noise covariance and the beamforming covariance may be determined based on the echo-cancelled input results.
- the noise covariance may be determined according to a larger value among the variance and the first constant value. Also, the noise covariance may be normalized according to a larger value among the variance and the first constant value.
- the beamforming covariance may be determined according to a larger value of the variance and the second constant value.
- the target signal extractor may repeatedly operate the echo canceller, the direction vector predictor, and the beamformer until the echo cancellation filter and the beamforming weight converge.
- the target signal extraction system may include an echo canceller, a direction vector predictor, and a beamformer.
- the echo canceller may include a weighted covariance generator, a weighted correlation vector generator, an echo cancellation filter generator, and an echo canceled signal generator.
- the echo canceller generates a weighted covariance based on a variance determined according to past input results for each frequency over time and output results corresponding to the echo-cancelled input results, and the input results for each frequency over time and generate a weighted correlation vector based on the variance determined according to output results corresponding to the past input results and the echo-cancelled input results, and echo based on the weighted covariance and the weighted correlation vector
- a cancellation filter may be generated, and the echo-cancelled input results may be generated based on the input results, the past input results, and the echo cancellation filter.
- the direction vector predictor generates an input signal covariance according to the echo-cancelled input results for each frequency over time, and generates noise based on a predetermined mask and a variance determined according to the output results corresponding to the input results.
- a covariance may be generated, and a direction vector may be generated based on the input signal covariance and the noise covariance.
- the beamformer generates a beamforming weight according to the echo-cancelled input results, a beamforming covariance determined according to the variance, and the direction vector, and based on the echo-cancelled input results and the beamforming weight, the Output results can be provided.
- the initial values of the noise covariance and the beamforming covariance may be determined according to a product of the echo-cancelled input results and the mask.
- the deechoed input results of the noise covariance may be updated as a product of the deechoed input results and the mask.
- the mask may be calculated for each frame index and frequency index.
- the noise covariance may be determined according to a larger value of the variance and a first constant value, and the noise covariance may be normalized according to a larger value of the variance and the first constant value. have.
- the beamforming covariance is determined according to a larger value of the variance and a second constant value
- the target signal extraction system includes the echo canceller until the echo cancellation filter and the beamforming weight converge;
- the direction vector predictor and the beamformer may be repeatedly operated.
- the online target signal extraction apparatus may include an echo canceller, a direction vector predictor, and a beamformer.
- the echo canceller may include a gain vector generator, a weighted inverse covariance generator, an echo cancellation filter generator, and a cancellation signal generator.
- the echo canceller generates a current frame echo cancellation output estimate based on current frame input results corresponding to the current frame, current frame past input results, and a previous frame echo canceling filter corresponding to the previous frame, and corresponding to the previous frame.
- a current frame echo cancellation variance estimate is generated based on the previous frame variance and the current frame echo cancellation output estimation value, and the previous frame weighted inverse covariance corresponding to the previous frame, the current frame echo cancellation output estimation value, and the current frame past input result generating a current frame gain vector based on , generate a current frame echo cancellation filter corresponding to the current frame based on the current frame past input results and the previous frame echo cancellation filter corresponding to the previous frame, and generate the current frame input results and the current frame past input results and the current frame echo cancellation filter, it is possible to generate current frame echo cancellation input results.
- the direction vector predictor generates a current frame input signal covariance generated based on the previous frame input signal covariance corresponding to the previous frame and the current frame echo cancellation input results for each frequency according to the current frame, and the current frame echo cancellation Generates a current frame variance estimate based on the input results and the previous frame beamforming weight, generates a current frame noise covariance based on a previous frame noise covariance corresponding to the previous frame and the current frame variance estimate value, and the current frame A current frame direction vector may be generated based on the frame input signal covariance, the current frame noise covariance, and the previous frame direction vector.
- the beamformer generates a current frame beamforming variance estimation value according to a previous frame beamforming weight, the current frame echo-removed input results, and the previous frame variance, and generates a previous frame inverse covariance and the current frame echo-rejected input results. and generating a current frame beamforming inverse covariance based on the current frame beamforming variance estimation value, generating a current frame beamforming weight according to the current frame beamforming inverse covariance and the current frame direction vector, and removing the current frame echo Current frame output results may be provided based on the received input results and the current frame beamforming weight.
- the current frame noise covariance may be normalized by the current frame variance estimate value.
- the on-line target signal extraction apparatus generates the current frame gain vector based on the current frame variance estimation value determined according to the current frame output results corresponding to the current frame input results.
- the target by calculating the current frame echo cancellation filter to generate the current frame echo cancellation input results, calculating the current frame noise covariance to generate the current frame direction vector, and updating the current frame beamforming weight It is possible to increase the extraction performance of the sound source.
- the online target signal extraction system may include an echo canceller, a direction vector predictor, and a beamformer.
- the echo canceller may include a gain vector generator, a weighted inverse covariance generator, an echo cancellation filter generator, and a cancellation signal generator.
- the echo canceller generates a current frame echo cancellation output estimate based on current frame input results corresponding to the current frame, current frame past input results, and a previous frame echo canceling filter corresponding to the previous frame, and corresponding to the previous frame. Generates a current frame echo cancellation variance estimate based on the previous frame variance and the echo cancellation output estimate, and adds the previous frame weighted inverse covariance corresponding to the previous frame, the current frame echo cancellation output estimate, and the current frame past input results.
- a current frame echo cancellation filter corresponding to the current frame is generated based on the current frame past input results and the previous frame echo cancellation filter corresponding to the previous frame, and the current frame input results and the current frame past input results and generating echo-cancelled input results of the current frame based on the current frame echo cancellation filter.
- the direction vector predictor generates the current frame input signal covariance corresponding to the previous frame and the current frame input signal covariance generated based on the current frame echo-removed input results for each frequency according to the current frame and the previous frame input signal covariance corresponding to the previous frame.
- a current frame noise covariance is generated based on a previous frame noise covariance, the current frame echo-cancelled input results, and a current frame variance estimate generated through a predetermined mask, the current frame input signal covariance, the current frame noise covariance and A current frame direction vector may be generated based on the previous frame direction vector.
- the beamformer generates a current frame beamforming variance estimate according to a previous frame beamforming weight, the current frame echo-removed input results, the previous frame variance, and the predetermined mask, and generates a previous frame inverse covariance and the current frame echo
- a current frame beamforming inverse covariance determined according to the removed input results and the current frame beamforming variance estimation value is generated, and a current frame beamforming weight is generated according to the current frame direction vector and the current frame inverse covariance, and , may provide current frame output results based on the current frame echo-cancelled input results and the current frame beamforming weight.
- the current frame noise covariance may be generated based on the previous frame noise covariance, the current frame echo-cancelled input results, and the current frame variance estimate generated through the predetermined mask.
- the current frame beamforming variance estimation value may be generated based on the previous frame beamforming weight, the current frame echo-cancelled input results, the previous frame variance, and the predetermined mask.
- the target signal extraction apparatus calculates the noise covariance based on the variance determined according to the output results corresponding to the input results, generates a direction vector, and updates the beamforming weight to improve the extraction performance for the target sound source. can be raised
- FIG. 1 is a view showing an apparatus for extracting a target signal according to embodiments of the present invention.
- FIG. 2 is a diagram illustrating an example of a direction vector predictor included in the apparatus for extracting a target signal of FIG. 1 .
- FIG. 3 is a diagram illustrating an example of a beamformer included in the apparatus for extracting a target signal of FIG. 1 .
- FIG. 4 is a diagram illustrating a target signal extraction system according to embodiments of the present invention.
- FIG. 5 is a diagram illustrating an example of a direction vector predictor included in the target signal extraction system of FIG. 4 .
- FIG. 6 is a diagram illustrating an example of a beamformer included in the target signal extraction system of FIG. 4 .
- FIG. 7 is a diagram illustrating an on-line target signal extraction apparatus according to embodiments of the present invention.
- FIG. 8 is a diagram illustrating an example of a direction vector predictor included in the online target signal extraction apparatus of FIG. 7 .
- FIG. 9 is a diagram illustrating an example of a beamformer included in the online target signal extraction apparatus of FIG. 7 .
- FIG. 10 is a diagram illustrating an online target signal extraction system according to embodiments of the present invention.
- FIG. 11 is a diagram illustrating an example of a direction vector predictor included in the online target signal extraction system of FIG. 10 .
- FIG. 12 is a diagram illustrating an example of a beamformer included in the online target signal extraction system of FIG. 10 .
- FIG. 13 is a diagram illustrating an example of an apparatus for extracting a target signal according to embodiments of the present invention.
- FIG. 14 is a diagram illustrating an example of an echo canceller included in the apparatus for extracting a target signal of FIG. 13 .
- 15 is a diagram illustrating an example of a direction vector predictor included in the apparatus for extracting a target signal of FIG. 13 .
- FIG. 16 is a diagram illustrating an example of a beamformer included in the apparatus for extracting a target signal of FIG. 13 .
- 17 is a diagram illustrating an example of a target signal extraction system according to embodiments of the present invention.
- FIG. 18 is a diagram illustrating an example of a direction vector predictor included in the target signal extraction system of FIG. 17 .
- FIG. 19 is a diagram illustrating an example of a beamformer included in the target signal extraction system of FIG. 17 .
- 20 is a diagram illustrating an example of an on-line target signal extraction apparatus according to embodiments of the present invention.
- 21 is a diagram illustrating an example of an echo canceller included in the online target signal extraction apparatus of FIG. 20 .
- FIG. 22 is a diagram illustrating an example of a direction vector predictor included in the online target signal extraction apparatus of FIG. 20 .
- FIG. 23 is a diagram illustrating an example of a beamformer included in the online target signal extraction apparatus of FIG. 20 .
- 24 is a diagram illustrating an example of an online target signal extraction system according to embodiments of the present invention.
- 25 is a diagram illustrating an example of a direction vector predictor included in the online target signal extraction system of FIG. 24 .
- 26 is a diagram illustrating an example of a beamformer included in the online target signal extraction system of FIG. 24 .
- FIG. 1 is a view showing an apparatus for extracting a target signal according to embodiments of the present invention
- FIG. 2 is a view showing an example of a direction vector predictor included in the apparatus for extracting a target signal of FIG. 1
- FIG. It is a view showing an example of a beamformer included in the target signal extraction apparatus.
- the target signal extraction apparatus 10 may include a direction vector predictor 100 and a beamformer 200 .
- the direction vector predictor 100 may include an input signal covariance generator 110 , a noise covariance generator 120 , and a vector generator 130 .
- the direction vector predictor 100 generates an input signal covariance (IC) according to the input results (XS) for each frequency according to time, and a variance determined according to the output results (OR) corresponding to the input results (XS)
- a noise covariance (NC) may be generated based on
- a direction vector (HV) may be generated based on the input signal covariance (IC) and the noise covariance (NC).
- the input signal covariance generator 110 may generate the input signal covariance IC according to the input results XS for each frequency over time.
- the input signal covariance (IC) can be expressed as [Equation 1] below.
- the input signal covariance is the number of frames, l is the frame index, k is the frequency index, may be input results.
- the noise covariance generator 120 may generate the noise covariance NC based on a variance determined according to the output results OR corresponding to the input results XS.
- the noise covariance (NC) can be expressed as [Equation 2] below.
- the noise covariance is distributed, is the first constant value, is the number of frames, l is the frame index, k is the frequency index, may be input results.
- the vector generator 130 may generate the direction vector HV based on the input signal covariance (IC) and the noise covariance (NC).
- the direction vector HV can be expressed as [Equation 3] below.
- the target sound source covariance is the eigenvector extraction function corresponding to the largest eigenvalue, may be a direction vector.
- the beamformer 200 generates a beamforming weight (BFW) according to a beamforming covariance (BC) and a direction vector (HV) determined according to the input results (XS) and the variance, and the input results (XS) and Output results OR may be provided based on the beamforming weight BFW.
- BFW beamforming weight
- BC beamforming covariance
- HV direction vector
- the beamformer 200 may include a beamforming weight generator 210 and an output generator 220 .
- the beamforming weight generator 210 may generate the beamforming weight BFW according to the input results XS and the beamforming covariance BC and the direction vector HV determined according to the variance.
- the beamforming covariance (BC) can be expressed as [Equation 4] below.
- the beamforming covariance may be a second constant value.
- the beamforming weight (BFW) can be expressed as in [Equation 5] below.
- the beamforming weight is the diagonal loading constant value, may be an identity matrix.
- the output generator 220 may provide output results OR based on the input results XS and the beamforming weight BFW.
- the variance of the noise covariance NC and the beamforming covariance BC may be determined based on the output results OR.
- the variance of the noise covariance (NC) and the beamforming covariance (BC) may be expressed as in [Equation 7] below.
- the output results may be the number of adjacent frames.
- initial values of the noise covariance NC and the beamforming covariance BC may be determined based on the input results XS.
- an initial value of variance used in noise covariance (NC) and beamforming covariance (BC) may be expressed as in [Equation 8] below.
- the noise covariance NC may be determined according to a larger value of the variance and the first constant value. Also, the noise covariance NC may be normalized according to a larger value among the variance and the first constant value. For example, the first constant value may be 10 ⁇ -6.
- the beamforming covariance BC may be determined according to a larger value among the variance and the second constant value.
- the second constant value may be 10 ⁇ -6.
- the target signal extraction apparatus 10 may repeatedly operate the direction vector predictor 100 and the beamformer 200 until the beamforming weight BFW converges. After generating the direction vector HV through the direction vector predictor 100 , the target signal extraction apparatus 10 may repeat the operation of generating the beamforming weight BFW through the beamformer 200 .
- the target signal extraction apparatus 10 according to the present invention generates a direction vector HV by calculating a noise covariance NC based on a variance determined according to the output results OR corresponding to the input results XS. And, by updating the beamforming weight (BFW), it is possible to increase the extraction performance for the target sound source.
- FIG. 4 is a diagram illustrating a target signal extraction system according to embodiments of the present invention
- FIG. 5 is a diagram illustrating an example of a direction vector predictor included in the target signal extraction system of FIG. 4, and
- FIG. It is a diagram showing an example of a beamformer included in the target signal extraction system.
- the target signal extraction system 11 may include a direction vector predictor 100 and a beamformer 200 .
- the direction vector predictor 100 may include an input signal covariance generator 110 , a noise covariance generator 120 , and a vector generator 130 .
- the direction vector predictor 100 generates an input signal covariance (IC) according to the input results (XS) for each frequency according to time, and a variance determined according to the output results (OR) corresponding to the input results (XS) and generating the noise covariance NC based on the predetermined mask MSK, and generating the direction vector HV based on the input signal covariance IC and the noise covariance NC.
- the beamformer 200 generates a beamforming weight (BFW) according to a beamforming covariance (BC) and a direction vector (HV) determined according to the input results (XS) and the variance, and the input results (XS) and Output results OR may be provided based on the beamforming weight BFW.
- BFW beamforming weight
- BC beamforming covariance
- HV direction vector
- the initial values of the noise covariance NC and the beamforming covariance may be determined according to a product of the input results XS and the mask MSK.
- the initial value of the variance used in the noise covariance (NC) can be expressed as [Equation 9] below.
- the input results XS of the noise covariance NC may be updated as a product of the input results XS and the mask MSK.
- the input results XS used in the noise covariance NC may be updated as in Equation 10 below.
- the mask MSK may be calculated for each frame index and frequency index.
- a mask for each frame index and frequency index may be calculated based on a neural network or diffusion.
- the noise covariance NC is determined according to a larger value of the variance and the first constant value, and the noise covariance NC is to be normalized according to the larger value of the variance and the first constant value.
- the beamforming covariance (BC) is determined according to the larger of the variance and the second constant value, and the target signal extraction system 11 operates the direction vector predictor ( 100) and the beamformer 200 may be repeatedly operated.
- FIG. 7 is a diagram illustrating an online target signal extraction apparatus according to embodiments of the present invention
- FIG. 8 is a diagram illustrating an example of a direction vector predictor included in the online target signal extraction apparatus of FIG. 7
- FIG. 9 is FIG. 7 is a diagram illustrating an example of a beamformer included in the online target signal extraction apparatus.
- the online target signal extraction apparatus 20 may include a direction vector predictor 100 and a beamformer 200 .
- the direction vector predictor 100 may include an input signal covariance generator 110 , a noise covariance generator 120 , and a vector generator 130 .
- the direction vector predictor 100 generates the current frame input signal covariance (C_IC) generated based on the previous frame input signal covariance (P_IC) corresponding to the previous frame and the current frame input results (C_XS) for each frequency according to the current frame.
- C_IC current frame input signal covariance
- P_IC previous frame input signal covariance
- C_XS current frame input results
- C_XS current frame input results
- P_BFW previous frame beamforming weight
- P_NC previous frame noise covariance
- P_NC current frame variance estimate corresponding to the previous frame
- C_NC current frame noise covariance
- C_HV current frame direction vector
- C_IC current frame input signal covariance
- C_NC current frame noise covariance
- P_HV previous frame direction vector
- the input signal covariance generator 110 generates a current frame input signal covariance generated based on a previous frame input signal covariance (P_IC) corresponding to the previous frame and current frame input results (C_XS) for each frequency according to the current frame. (C_IC) can be created.
- P_IC previous frame input signal covariance
- C_XS current frame input results
- the current frame input signal covariance (C_IC) can be expressed as in [Equation 11] below.
- the noise covariance generator 120 includes the previous frame noise covariance (P_NC) corresponding to the previous frame and the current frame input results (C_XS) for each frequency and the previous frame beamforming weight (P_BFW) corresponding to the input results in the previous frame.
- the current frame noise covariance (C_NC) may be generated based on the current frame variance estimate generated according to .
- C_NC current frame noise covariance
- the current frame noise covariance is the current frame noise covariance, silver forgetting factor, is the previous frame noise covariance, is the current frame variance estimate, is the current frame estimation output results, is the beamforming weight of the previous frame, is the current frame input results, may be a third constant value.
- the vector generator 130 may generate the current frame direction vector C_HV based on the current frame input signal covariance C_IC and the current frame noise covariance C_NC.
- the current frame direction vector (C_HV) can be expressed as in [Equation 13] below.
- the current frame direction vector is the previous frame direction vector, is the current frame target sound source covariance, is the normalized current frame direction vector, may be one element of the normalized current frame direction vector.
- the beamformer 200 generates a current frame beamforming variance estimation value according to the previous frame beamforming weight (P_BFW), the current frame input results (C_XS), and the previous frame variance (P_V), and the previous frame inverse covariance (P_IBC) , the current frame input results (C_XS), the current frame beamforming inverse covariance (C_IBC) is generated based on the current frame beamforming variance estimation value, and the current frame beamforming inverse covariance (C_IBC) and the current frame direction vector (C_HV) are Accordingly, the current frame beamforming weight C_BFW may be generated, and current frame output results C_OR may be provided based on the current frame input results C_XS and the current frame beamforming weight C_BFW.
- the beamformer 200 may include a beamforming weight generator 210 and an output generator 220 .
- the beamforming weight generator 210 generates a current frame beamforming variance estimation value according to the current frame input results (C_XS), the previous frame beamforming weight (P_BFW), and the previous frame variance (P_V), and the current frame input results ( C_XS), the previous frame beamforming inverse covariance (P_IBC), and the current frame beamforming variance estimate value to generate the current frame beamforming inverse covariance (C_IBC), and the current frame beamforming inverse covariance (C_IBC) and the current frame direction vector (C_HV). ), the current frame beamforming weight (C_BFW) may be generated.
- the current frame beamforming variance estimation value can be expressed as [Equation 14] below.
- the current frame beamforming variance estimate is the current frame estimation output results, is the previous frame distribution, is the weight, may be a fourth constant value.
- the current frame beamforming weight (C_BFW) can be expressed as in [Equation 15] below.
- the current frame beamforming weight is the previous frame beamforming inverse covariance, is the current frame direction vector, may be the current frame beamforming inverse covariance.
- the output generator 220 may provide the current frame output results C_OR based on the current frame input results C_XS and the current frame beamforming weight C_BFW.
- the current frame noise covariance (C_NC) may be normalized by the current frame variance estimation value.
- the on-line target signal extraction apparatus 20 calculates the current frame noise covariance based on the current frame variance estimation value determined according to the current frame output results C_OR corresponding to the current frame input results C_XS.
- C_HV current frame direction vector
- C_BFW current frame beamforming weight
- FIG. 10 to 12 are diagrams illustrating an online target signal extraction system according to embodiments of the present invention
- FIG. 11 is a diagram illustrating an example of a direction vector predictor included in the online target signal extraction system of FIG.
- FIG. 10 is a diagram illustrating an example of a beamformer included in the online target signal extraction system of FIG. 10 .
- the online target signal extraction system 21 may include a direction vector predictor 100 and a beamformer 200 .
- the direction vector predictor 100 may include an input signal covariance generator 110 , a noise covariance generator 120 , and a vector generator 130 .
- the direction vector predictor 100 generates the current frame input signal covariance (C_IC) generated based on the previous frame input signal covariance (P_IC) corresponding to the previous frame and the current frame input results (C_XS) for each frequency according to the current frame.
- the current frame direction vector C_HV may be generated based on the frame input signal covariance (C_IC), the current frame noise covariance (C_NC), and the previous frame direction vector (P_HV).
- the beamformer 200 generates a current frame beamforming variance estimate according to the previous frame beamforming weight (P_BFW), the current frame input results (C_XS), the previous frame variance, and a predetermined mask, and the previous frame inverse covariance (P_IBC). ), the current frame input results (C_XS) and the current frame beamforming inverse covariance (C_IBC) determined according to the current frame beamforming variance estimation value, and the current frame direction vector (C_HV) and the current frame beamforming inverse covariance (C_IBC) ), the current frame beamforming weight C_BFW may be generated, and current frame output results C_OR may be provided based on the current frame input results C_XS and the current frame beamforming weight C_BFW.
- the current frame noise covariance (C_NC) may be generated based on the previous frame noise covariance (P_NC), the current frame input results (C_XS), and a current frame variance estimate generated through a predetermined mask.
- the current frame noise covariance (C_NC) can be expressed as in [Equation 17] below.
- the current frame noise covariance is a mask, silver forgetting factor, is the previous frame noise covariance, is the current frame variance estimate, is the element of the input results of the current frame, may be a third constant value.
- the current frame beamforming variance estimation value may be generated based on the previous frame beamforming weight (P_BFW), the current frame input results (C_XS), the previous frame variance (P_V), and a predetermined mask.
- the current frame beamforming variance estimate may be expressed as in [Equation 18] below.
- the current frame estimation output results is the beamforming weight of the previous frame, is the current frame input results, is a mask, is the current frame beamforming variance estimate, is the previous frame distribution, is the weight, may be a fourth constant value.
- FIG. 13 to 16 are diagrams illustrating examples of a target signal extraction apparatus according to embodiments of the present invention
- FIG. 14 is a diagram illustrating an example of an echo canceller included in the target signal extraction apparatus of FIG. 13
- FIG. 15 is 13 is a diagram illustrating an example of a direction vector predictor included in the apparatus for extracting a target signal of FIG. 13
- FIG. 16 is a diagram illustrating an example of a beamformer included in the apparatus for extracting a target signal of FIG. 13 .
- the target signal extraction apparatus 30 may include an echo canceller 300 , a direction vector predictor 100 , and a beamformer 200 .
- the echo canceller 300 may include a weighted covariance generator 310 , a weighted correlation vector generator 320 , a de-echo filter generator 330 , and a de-echo-cancelled signal generator 340 .
- the echo canceller 300 calculates the weighted covariance WC based on the variance determined according to the output results OR corresponding to the past input results XPS and the echo-cancelled input results DS for each frequency over time.
- the echo-cancelled input results DS may be generated based on the results XPS and the echo cancellation filter DF.
- the weighted covariance generator 310 may generate the weighted covariance WC according to the past input results XPS and the variance.
- the weighted covariance (WC) can be expressed as [Equation 19] below.
- weighted covariance is the past input results
- silver dispersion is the number of delay frames, number of taps, may be a second constant value.
- the weighted correlation vector generator 320 may generate the correlation vector WV weighted according to the frequency-dependent input results XS, past input results, and variance according to time.
- the weighted correlation vector WV can be expressed as [Equation 20] below.
- weighted correlation vector may be current frame input results.
- the de-echo filter generator 330 may generate the de-echo filter DF based on the weighted covariance (WC) and the weighted correlation vector (WV).
- the echo cancellation filter DF may be expressed as in [Equation 21] below.
- the echo-cancelled signal generator 340 may generate the echo-cancelled input results DS based on the input results XS, the past input results XPS, and the echo cancellation filter DF.
- the echo-cancelled input results DS can be expressed as in [Equation 22] below.
- the direction vector predictor 100 generates an input signal covariance IC according to the echo-cancelled input results DS, and based on the variance determined according to the output results OR corresponding to the input results XS to generate a noise covariance (NC), and a direction vector (HV) may be generated based on the input signal covariance (IC) and the noise covariance (NC).
- the input signal covariance generator 110 may generate the input signal covariance IC according to the echo-cancelled input results DS.
- the input signal covariance (IC) can be expressed as [Equation 23] below.
- the input signal covariance is the number of frames
- l is the frame index
- k is the frequency index
- the noise covariance generator 120 may generate the noise covariance NC based on a variance determined according to output results OR corresponding to the echo-cancelled input results DS.
- the noise covariance (NC) can be expressed as [Equation 24] below.
- noise covariance is distributed, is the first constant value, is the number of frames, l is the frame index, k is the frequency index, may be echo-cancelled input results.
- the vector generator 130 may generate the direction vector HV based on the input signal covariance (IC) and the noise covariance (NC). For example, the content of [Equation 3] described with reference to FIGS. 1 to 3 may be equally applied to the direction vector HV.
- the beamformer 200 generates a beamforming weight (BFW) according to the echo-cancelled input results (DS), the beamforming covariance (BS) determined according to the variance, and the direction vector (HV), and generates the echo-cancelled input.
- Output results OR may be provided based on the results DS and the beamforming weight BFW.
- the beamformer 200 may include a beamforming weight generator 210 and an output generator 220 .
- the beamforming weight generator 210 may generate the beamforming weight BFW according to the echo-cancelled input results DS, the beamforming covariance BC determined according to the variance, and the direction vector HV.
- the beamforming covariance (BC) can be expressed as [Equation 25] below.
- the beamforming covariance may be a second constant value.
- the beamforming weight (BFW) can be expressed as in [Equation 26] below.
- the beamforming weight is the diagonal loading constant value, may be an identity matrix.
- the output generator 220 may provide output results OR based on the echo-cancelled input results DS and the beamforming weight BFW.
- the weighted covariance (WC), the weighted correlation vector (WV), and the noise covariance (NC) and the beamforming covariance (BC) may be determined based on the output results (OR).
- the weighted covariance (WC) and the variance used in the weighted correlation vector (WV) are equally applicable to the contents of [Equation 7] described in FIGS. 1 to 3 .
- initial values of the weighted covariance WC and the weighted correlation vector WV may be determined based on the input results XS.
- the initial value of the variance used in the weighted covariance (WC) and the weighted correlation vector (WV) may be expressed as [Equation 28] below.
- the number of adjacent frames is the number of channels of the input results, may be a frame index.
- the weighted covariance WC and the weighted correlation vector WV may be determined according to a larger value of the variance and the second constant value.
- initial values of the noise covariance NC and the beamforming covariance BC may be determined based on the echo-cancelled input results DS.
- an initial value of variance used in noise covariance (NC) and beamforming covariance (BC) may be expressed as [Equation 29] below.
- the noise covariance NC may be determined according to a larger value of the variance and the first constant value. Also, the noise covariance NC may be normalized according to a larger value among the variance and the first constant value.
- the beamforming covariance BC may be determined according to a larger value among the variance and the second constant value.
- the target signal extraction apparatus 30 includes the echo canceller 300, the direction vector predictor 100, and the beamformer 200 until the echo cancellation filter DF and the beamforming weight BFW converge. ) can be repeatedly operated.
- the target signal extraction apparatus 30 generates the echo-cancelled input results DS through the echo canceller 300 , and generates the direction vector HV through the direction vector predictor 100 , and then generates a beamformer ( 200), the operation of generating the beamforming weight (BFW) may be repeated.
- the target signal extraction apparatus 30 according to the present invention calculates the weighted covariance WC and the weighted correlation vector WV based on the variance determined according to the output results OR corresponding to the input results XS.
- the target by calculating the dereverberation filter (DF) through the It is possible to increase the extraction performance of the sound source.
- FIG. 17 to 19 are diagrams illustrating examples of a target signal extraction system according to embodiments of the present invention
- FIG. 18 is a diagram illustrating an example of a direction vector predictor included in the target signal extraction system of FIG. 17,
- FIG. 19 FIG. 17 is a diagram illustrating an example of a beamformer included in the target signal extraction system of FIG. 17 .
- the target signal extraction system 31 may include an echo canceller 300 , a direction vector predictor 100 , and a beamformer 200 .
- the echo canceller 300 may include a weighted covariance generator 310 , a weighted correlation vector generator 320 , a de-echo filter generator 330 , and a de-echo-cancelled signal generator 340 .
- the echo canceller 300 calculates the weighted covariance WC based on the variance determined according to the output results OR corresponding to the past input results XPS and the echo-cancelled input results DS for each frequency over time.
- the echo-cancelled input results DS may be generated based on the results XPS and the echo cancellation filter DF.
- the direction vector predictor 100 generates an input signal covariance (IC) according to the echo-cancelled input results DS for each frequency according to time, and generates output results OR corresponding to the echo-cancelled input results DS. ) and generates a noise covariance (NC) based on a predetermined mask (MSK) and a direction vector (HV) based on the input signal covariance (IC) and noise covariance (NC). .
- the beamformer 200 generates a beamforming weight (BFW) according to the dereflected input results (DS) and the beamforming covariance (BC) and the direction vector (HV) determined according to the variance, and generates the dereverberated input. Output results OR may be provided based on the results DS and the beamforming weight BFW.
- initial values of the noise covariance NC and the beamforming covariance BC may be determined according to a product of the echo-cancelled input results DS and the mask MSK.
- an initial value of variance used in noise covariance (NC) and beamforming covariance (BC) may be expressed as in [Equation 30] below.
- silver mask is the echo-cancelled input results, may be the number of channels of input results.
- the de-echoed input results DS of the noise covariance NC may be updated as a product of the de-echoed input results DS and the mask MSK.
- the echo-cancelled input results DS used in the noise covariance NC may be updated as in [Equation 31] below.
- the mask MSK may be calculated for each frame index and frequency index.
- a mask for each frame index and frequency index may be calculated based on a neural network or diffusion.
- the noise covariance NC is determined according to a larger value of the variance and the first constant value, and the noise covariance NC is to be normalized according to the larger value of the variance and the first constant value.
- the beamforming covariance BC is determined according to a larger value among the variance and the second constant value, and the target signal extraction system 31 converges the echo cancellation filter DF and the beamforming weight BFW.
- the echo canceller 300 , the direction vector predictor 100 , and the beamformer 200 may be repeatedly operated until .
- FIG. 20 to 23 are diagrams illustrating examples of an on-line target signal extraction apparatus according to embodiments of the present invention
- FIG. 21 is a diagram illustrating an example of an echo canceller included in the online target signal extraction apparatus of FIG. 20
- 22 is a diagram illustrating an example of a direction vector predictor included in the online target signal extraction apparatus of FIG. 20
- FIG. 23 is a diagram illustrating an example of a beamformer included in the online target signal extraction apparatus of FIG. 20 .
- the online target signal extraction apparatus 40 may include an echo canceller 300 , a direction vector predictor 100 , and a beamformer 200 .
- the echo canceller 300 may include a gain vector generator 350 , a weighted inverse covariance generator 360 , a cancellation filter generator 330 , and a cancellation signal generator 340 .
- the echo canceller 300 cancels echo of the current frame based on the current frame input results (C_XS) corresponding to the current frame, the current frame past input results (C_XPS), and the previous frame echo cancellation filter (P_DF) corresponding to the previous frame.
- C_EDS Generates an output estimation value (C_EDS), generates a current frame echo cancellation variance estimation value based on a previous frame variance (P_V) corresponding to the previous frame and a current frame echo cancellation output estimation value (C_EDS) corresponding to the previous frame, and generates a previous frame corresponding to the previous frame
- the current frame gain vector (C_GV) is generated based on the weighted inverse covariance (P_IWC), the current frame echo cancellation output estimate (C_EDS), and the current frame past input results (C_XPS), and the previous frame weighted inverse covariance (P_IWC) , generates the current frame weighted inverse covariance (C_IWC) based on the current frame past input results (C_XPS) and the current frame gain vector (C_GV), and the current frame gain vector (C_GV) and the current frame past input results (C_XPS) ) and the previous frame echo cancellation filter (P_DF) corresponding to the previous frame, the current frame
- the gain vector generator 350 calculates the current frame echo cancellation output estimation value (C_EDS) based on the current frame input results (C_XS), the current frame past input results (C_XPS), and the previous frame echo cancellation filter (P_DF).
- C_EDS current frame echo cancellation output estimation value
- C_XS current frame input results
- C_XPS current frame past input results
- P_DF previous frame echo cancellation filter
- C_EDS The current frame echo cancellation output estimation value (C_EDS) can be expressed as in [Equation 32] below.
- the current frame echo cancellation output estimate is the current frame input results, is the previous frame echo cancellation filter, may be input results in the past of the current frame.
- the gain vector generator 350 may generate an estimated current frame de-echo-cancellation variance based on the previous frame variance (P_V) and the current frame de-echo-cancellation output estimate (C_EDS).
- the current frame echo cancellation variance estimate can be expressed as [Equation 33] below.
- the gain vector generator 350 may generate the current frame gain vector C_GV based on the previous frame weighted inverse covariance P_IWC, the current frame past input results C_XPS, and the current frame variance estimate value.
- the current frame gain vector (C_GV) can be expressed as in [Equation 34] below.
- the weighted inverse covariance generator 360 generates the current frame weighted inverse covariance (C_IWC) based on the previous frame weighted inverse covariance (P_IWC), the current frame past input results (P_XPS), and the current frame gain vector (C_GV). can do.
- C_IWC current frame weighted inverse covariance
- the echo cancellation filter generator 330 generates a current frame echo cancellation filter (C_DF) based on the previous frame echo cancellation filter (P_DF), the current frame echo cancellation output estimate value (C_EDS), and the current frame past input results (C_XPS).
- C_DF current frame echo cancellation filter
- P_DF previous frame echo cancellation filter
- C_EDS current frame echo cancellation output estimate value
- C_XPS current frame past input results
- the current frame echo cancellation filter (C_DF) can be expressed as [Equation 36] below.
- the current frame echo cancellation filter is the previous frame echo cancellation filter, is the current frame gain vector, may be a current frame echo cancellation output estimate value.
- the echo cancellation signal generator 340 generates the current frame echo cancellation input results C_DS based on the current frame input results C_XS, the current frame echo cancellation filter C_DF, and the current frame past input results C_XPS. can create
- the current frame echo cancellation input results C_DS can be expressed as in [Equation 37] below.
- the current frame echo-removed input results may be a current frame echo cancellation filter.
- the direction vector predictor 100 is a current frame input signal covariance (C_IC) generated based on the previous frame input signal covariance (P_IC) corresponding to the previous frame and the current frame echo-cancelled input results (C_DS) for each frequency according to the current frame ), generate a current frame variance estimate based on the current frame echo-cancelled input results (C_DS) and the previous frame beamforming weight (P_BFW), and the previous frame noise covariance (P_NC) and current
- the current frame noise covariance (C_NC) is generated based on the frame variance estimate, and the current frame direction vector (C_HV) is based on the current frame input signal covariance (C_IC), the current frame noise covariance (C_NC), and the previous frame direction vector (P_HV). ) can be created.
- the input signal covariance generator 110 generates the current frame based on the previous frame input signal covariance (P_IC) corresponding to the previous frame and the current frame echo-cancelled input results (C_DS) for each frequency according to the current frame.
- An input signal covariance (C_IC) can be generated.
- the current frame input signal covariance (C_IC) can be expressed as in [Equation 38] below.
- the noise covariance generator 120 includes the previous frame noise covariance (P_NC) corresponding to the previous frame and the current frame echo-removed input results (C_DS) for each frequency and the previous frame beamforming weight corresponding to the input results in the previous frame.
- the current frame noise covariance (C_NC) may be generated based on the current frame variance estimate generated according to (P_BFW).
- C_NC current frame noise covariance
- the current frame noise covariance is the current frame noise covariance, silver forgetting factor, is the previous frame noise covariance, is the current frame variance estimate, is the current frame estimation output results, is the beamforming weight of the previous frame, is the current frame echo-removed input results, may be a third constant value.
- the vector generator 130 may generate the current frame direction vector (C_HV) based on the current frame input signal covariance (C_IC) and the current frame noise covariance (C_NC), and [Equation 13] can be equally applied.
- the beamformer 200 generates a current frame beamforming variance estimate according to the previous frame beamforming weight (P_BFW), the current frame echo-removed input results (C_DS), and the previous frame variance (P_V), and the previous frame inverse covariance (P_IBC), current frame echo-cancelled input results (C_DS), and current frame beamforming inverse covariance (C_IBC) are generated based on the current frame beamforming variance estimation value, and current frame beamforming inverse covariance (C_IBC) and current frame A current frame beamforming weight (C_BFW) is generated according to the direction vector (C_HV), and current frame output results (C_OR) are generated based on the current frame echo-removed input results (C_DS) and the current frame beamforming weight (C_BFW) can provide
- the beamformer 200 may include a beamforming weight generator 210 and an output generator 220 .
- the beamforming weight generator 210 generates a current frame beamforming variance estimate according to the current frame echo-removed input results (C_DS), the previous frame beamforming weight (P_BFW), and the previous frame variance (P_V), and the current frame echo
- a current frame beamforming inverse covariance (C_IBC) is generated through the removed input results (C_DS) and the previous frame beamforming inverse covariance (P_IBC) and the current frame beamforming variance estimation value, and the current frame beamforming inverse covariance (C_IBC) and A current frame beamforming weight (C_BFW) may be generated according to the current frame direction vector (C_HV).
- the current frame beamforming weight (C_BFW) can be expressed as in [Equation 40] below.
- the current frame beamforming weight is the previous frame beamforming inverse covariance
- the current frame direction vector is the current frame beamforming inverse covariance
- the current frame beamforming inverse covariance may be input results from which the echo of the current frame has been removed.
- the output generator 220 may provide the current frame output results C_OR based on the current frame echo-cancelled input results C_DS and the current frame beamforming weight C_BFW.
- the current frame noise covariance (C_NC) may be normalized by the current frame variance estimate value.
- the on-line target signal extraction apparatus 40 provides a current frame gain vector (C_GV) based on a current frame variance estimation value determined according to the current frame output results (C_OR) corresponding to the current frame input results (C_XS).
- C_DF current frame echo cancellation filter
- C_DS current frame echo cancellation input results
- C_NC current frame noise covariance
- C_HV current frame direction vector
- C_BFW beamforming weight
- FIG. 24 to 26 are diagrams illustrating an online target signal extraction system according to embodiments of the present invention
- FIG. 25 is a diagram illustrating an example of a direction vector predictor included in the online target signal extraction system of FIG. 24,
- FIG. 26 FIG. 24 is a diagram illustrating an example of a beamformer included in the online target signal extraction system of FIG. 24 .
- the online target signal extraction system 41 may include an echo canceller 300 , a direction vector predictor 100 , and a beamformer 200 .
- the echo canceller 300 may include a gain vector generator 350 , a weighted inverse covariance generator 360 , a cancellation filter generator 330 , and a cancellation signal generator 340 .
- the echo canceller 300 cancels echo of the current frame based on the current frame input results (C_XS) corresponding to the current frame, the current frame past input results (C_XPS), and the previous frame echo cancellation filter (P_DF) corresponding to the previous frame.
- C_EDS Generate an output estimate
- C_EDS Generate an output estimate
- C_EDS Generate an output estimate
- P_V previous frame variance
- C_EDS echo-cancellation output estimate
- C_EDS current frame gain vector
- P_IWC inverse covariance
- C_XPS current frame echo cancellation output estimate
- C_IWC current frame past input results
- P_IWC previous frame weighted inverse covariance
- the current A current frame weighted inverse covariance (C_IWC) is generated based on the frame past input results (C_XPS) and the current frame gain vector (C_GV), and the current frame gain vector (C_GV), the current frame past input results (C_XPS) and A current frame echo cancellation filter (C_DF) corresponding to the current frame is generated based on the previous frame echo cancellation filter (P_DF) corresponding to the previous frame echo cancellation filter (P_DF) corresponding to the previous frame echo cancellation filter (P_DF) corresponding to the previous frame echo cancellation filter
- the direction vector predictor 100 is a current frame input signal covariance (C_IC) generated based on the previous frame input signal covariance (P_IC) corresponding to the previous frame and the current frame echo-cancelled input results (C_DS) for each frequency according to the current frame ), and the current frame noise covariance (C_NC) based on the previous frame noise covariance (P_NC) corresponding to the previous frame, the current frame echo-cancelled input results (C_DS), and the current frame variance estimate generated through a predetermined mask. ) and generate a current frame direction vector (C_HV) based on the current frame input signal covariance (C_IC), the current frame noise covariance (C_NC), and the previous frame direction vector (P_HV).
- the beamformer 200 generates a current frame beamforming variance estimation value according to the previous frame beamforming weight (P_BFW), the current frame echo-removed input results (C_DS), the previous frame variance, and a predetermined mask, and inverses the previous frame
- a current frame beamforming inverse covariance (C_IBC) determined according to the covariance (P_IBC), the current frame echo-cancelled input results (C_DS), and the current frame beamforming variance estimation value is generated, and the current frame direction vector (C_HV) and the current frame
- the current frame beamforming weight (C_BFW) is generated according to the beamforming inverse covariance (C_IBC), and the current frame output results (C_BFW) based on the current frame echo-removed input results (C_DS) and the current frame beamforming weight (C_BFW) C_OR) can be provided.
- the target signal extraction system 41 according to the present invention is [Equation 13] to [Equation 14] described in FIGS. 7 to 9 and [Equation 32] to [Equation 37] described in FIGS. 20 to 23 and The contents of [Equation 39] may be equally applied.
- the current frame noise covariance (C_NC) is to be generated based on the previous frame noise covariance (P_NC) and the current frame echo-cancelled input results (C_DS) and the current frame variance estimate generated through a predetermined mask.
- the current frame noise covariance (C_NC) may be expressed as in [Equation 42] below.
- the current frame noise covariance is a mask, silver forgetting factor, is the previous frame noise covariance, is the current frame variance estimate, is the current frame echo-removed input results, may be a third constant value.
- the current frame beamforming variance estimation value may be generated based on the previous frame beamforming weight (P_BFW), the current frame echo-cancelled input results (C_DS), the previous frame variance (P_V), and a predetermined mask.
- P_BFW previous frame beamforming weight
- C_DS current frame echo-cancelled input results
- P_V previous frame variance
- the current frame beamforming variance estimate may be expressed as in [Equation 43] below.
- the current frame estimation output results is the beamforming weight of the previous frame, is the current frame echo-removed input results, is a mask, is the current frame beamforming variance estimate, is the previous frame distribution, is the weight, may be a fourth constant value.
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Circuit For Audible Band Transducer (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
Description
Claims (18)
- 시간에 따른 주파수별 입력결과들에 따라 입력신호 공분산을 생성하고, 상기 입력결과들에 상응하는 출력결과들에 따라 결정되는 분산에 기초하여 노이즈 공분산을 생성하며, 상기 입력신호 공분산 및 상기 노이즈 공분산에 기초하여 방향 벡터를 생성하는 방향 벡터 예측기; 및상기 분산에 따라 결정되는 빔포밍 공분산 및 상기 방향 벡터에 따라 빔포밍 가중치를 생성하고, 상기 입력결과들 및 상기 빔포밍 가중치에 기초하여 상기 출력결과들을 제공하는 빔포밍기를 포함하는 타겟신호 추출장치.
- 제1항에 있어서,상기 노이즈 공분산 및 빔포밍 공분산의 초기값은 상기 입력결과들에 기초하여 결정되는 것을 특징으로 하는 타겟신호 추출장치.
- 제2항에 있어서,상기 노이즈 공분산은 상기 분산 및 제1 상수값 중 큰 값에 따라 결정되는 것을 특징으로 하는 타겟신호 추출장치.
- 제3항에 있어서,상기 노이즈 공분산은 상기 분산 및 상기 제1 상수값 중 큰 값에 따라 노말라이제이션(Normalization)되는 것을 특징으로 하는 타겟신호 추출장치.
- 제4항에 있어서,상기 빔포밍 공분산은 상기 분산 및 제2 상수값 중 큰 값에 따라 결정되는 것을 특징으로 하는 타겟신호 추출장치.
- 제5항에 있어서,상기 타겟신호 추출장치는 상기 빔포밍 가중치가 수렴할 때까지 상기 방향 벡터 예측기 및 상기 빔포밍기를 반복적으로 동작시키는 것을 특징으로 하는 타겟신호 추출장치.
- 시간에 따른 주파수별 입력결과들에 따라 입력신호 공분산을 생성하고, 상기 입력결과들에 상응하는 출력결과들에 따라 결정되는 분산 및 미리 결정된 마스크에 기초하여 노이즈 공분산을 생성하며, 상기 입력신호 공분산 및 상기 노이즈 공분산에 기초하여 방향 벡터를 생성하는 방향 벡터 예측기; 및상기 분산에 따라 결정되는 빔포밍 공분산 및 상기 방향 벡터에 따라 빔포밍 가중치를 생성하고, 상기 입력결과들 및 상기 빔포밍 가중치에 기초하여 상기 출력결과들을 제공하는 빔포밍기를 포함하는 타겟신호 추출시스템.
- 제7항에 있어서,상기 노이즈 공분산의 초기값은 상기 입력결과들 및 상기 마스크의 곱에 따라 결정되는 것을 특징으로 하는 타겟신호 추출시스템.
- 제8항에 있어서,상기 노이즈 공분산은 상기 분산 및 제1 상수값 중 큰 값에 따라 결정되고,상기 노이즈 공분산은 상기 분산 및 상기 제1 상수값 중 큰 값에 따라 노말라이제이션(Normalization)되는 것을 특징으로 하는 타겟신호 추출시스템.
- 제9항에 있어서,상기 빔포밍 공분산은 상기 분산 및 제2 상수값 중 큰 값에 따라 결정되고,상기 타겟신호 추출장치는 상기 빔포밍 가중치가 수렴할 때까지 상기 방향 벡터 예측기 및 상기 빔포밍기를 반복적으로 동작시키는 것을 특징으로 하는 타겟신호 추출시스템.
- 이전프레임에 상응하는 이전프레임 입력신호 공분산 및 현재 프레임에 따른 주파수별 현재프레임 입력결과들에 기초하여 생성되는 현재프레임 입력신호 공분산을 생성하고, 상기 현재프레임 입력결과들 및 이전프레임 빔포밍 가중치에 기초하여 현재프레임 분산 추정값을 생성하고, 이전프레임에 상응하는 이전프레임 노이즈 공분산 및 현재프레임 분산 추정값에 기초하여 현재프레임 노이즈 공분산을 생성하며, 현재프레임 입력신호 공분산 및 현재프레임 노이즈 공분산 및 이전프레임 방향 벡터에 기초하여 현재프레임 방향 벡터를 생성하는 방향 벡터 예측기; 및상기 이전프레임 빔포밍 가중치, 상기 현재프레임 입력결과들, 이전프레임 분산에 따라 현재프레임 빔포밍 분산 추정값을 생성하고, 이전프레임 역 공분산, 상기 현재프레임 입력결과들, 상기 현재프레임 빔포밍 분산 추정값에 기초하여 현재프레임 빔포밍 역 공분산을 생성하고, 상기 현재프레임 빔포밍 역 공분산 및 상기 현재프레임 방향 벡터에 따라 현재프레임 빔포밍 가중치를 생성하고, 상기 현재프레임 입력결과들 및 상기 현재프레임 빔포밍 가중치에 기초하여 현재프레임 출력결과들을 제공하는 빔포밍기를 포함하는 온라인 타겟신호 추출장치.
- 제11항에 있어서,상기 현재프레임 노이즈 공분산은 현재프레임 분산 추정값에 의해 노말라이제이션(Normalization)되는 것을 특징으로 하는 온라인 타겟신호 추출장치.
- 이전프레임에 상응하는 이전프레임 입력신호 공분산 및 현재 프레임에 따른 주파수별 현재프레임 입력결과들에 기초하여 생성되는 현재프레임 입력신호 공분산을 생성하고, 상기 이전프레임에 상응하는 이전프레임 노이즈 공분산, 상기 현재프레임 입력결과들 및 미리 결정된 마스크를 통해 생성된 현재프레임 분산 추정값에 기초하여 현재프레임 노이즈 공분산을 생성하며, 상기 현재프레임 입력신호 공분산, 상기 현재프레임 노이즈 공분산 및 이전프레임 방향 벡터에 기초하여 현재프레임 방향 벡터를 생성하는 방향 벡터 예측기; 및이전프레임 빔포밍 가중치, 상기 현재프레임 입력결과들, 이전프레임 분산 및 상기 미리 결정된 마스크에 따라 현재프레임 빔포밍 분산 추정값을 생성하고, 이전프레임 역 공분산, 상기 현재프레임 입력결과들 및 상기 현재프레임 빔포밍 분산 추정값에 따라 결정되는 현재프레임 빔포밍 역 공분산을 생성하고, 상기 현재프레임 방향 벡터 및 상기 현재프레임 빔포밍 역 공분산에 따라 현재프레임 빔포밍 가중치를 생성하고, 상기 현재프레임 입력결과들 및 상기 현재프레임 빔포밍 가중치에 기초하여 현재프레임 출력결과들을 제공하는 빔포밍기를 포함하는 온라인 타겟신호 추출시스템.
- 시간에 따른 주파수별 과거 입력결과들, 입력결과들 및 출력결과들에 기초하여 반향 제거된 입력결과들을 제공하는 반향 제거기;상기 반향 제거된 입력결과들에 따라 입력신호 공분산을 생성하고, 상기 반향 제거된 입력결과들에 상응하는 출력결과들에 따라 결정되는 분산에 기초하여 노이즈 공분산을 생성하며, 상기 입력신호 공분산 및 상기 노이즈 공분산에 기초하여 방향 벡터를 생성하는 방향 벡터 예측기; 및상기 분산에 따라 결정되는 빔포밍 공분산 및 상기 방향 벡터에 따라 빔포밍 가중치를 생성하고, 상기 반향 제거된 입력결과들 및 상기 빔포밍 가중치에 기초하여 상기 출력결과들을 제공하는 빔포밍기를 포함하는 타겟신호 추출장치.
- 제14항에 있어서,상기 반향 제거기는,시간에 따른 주파수별 상기 과거 입력결과들 및 상기 반향 제거된 입력결과들에 상응하는 상기 출력결과들에 따라 결정되는 분산에 기초하여 가중된 공분산을 제공하는 가중된 공분산 생성기;상기 시간에 따른 주파수별 상기 입력결과들 및 상기 과거 입력결과들 및 상기 반향 제거된 입력결과들에 상응하는 상기 출력결과들에 따라 결정되는 분산에 기초하여 가중된 상관 벡터를 제공하는 가중된 상관 벡터 생성기;상기 가중된 공분산 및 상기 가중된 상관 벡터에 기초하여 반향 제거 필터를 생성하는 반향제거 필터 생성기; 및상기 입력결과들 및 상기 과거 입력결과들 및 상기 반향 제거 필터에 기초하여 상기 반향 제거된 입력결과들을 생성하는 반향 제거된 신호 생성기를 포함하는 것을 특징으로 하는 타겟신호 추출장치.
- 시간에 따른 주파수별 과거 입력결과들, 입력결과들 및 출력결과들에 기초하여 반향 제거된 입력결과들을 제공하는 반향 제거기;상기 반향 제거된 입력결과들에 따라 입력신호 공분산을 생성하고, 상기 반향 제거된 입력결과들에 상응하는 출력결과들에 따라 결정되는 분산 및 미리 결정된 마스크에 기초하여 노이즈 공분산을 생성하며, 상기 입력신호 공분산 및 노이즈 공분산에 기초하여 방향 벡터를 생성하는 방향 벡터 예측기; 및상기 반향 제거된 입력결과들 및 분산에 따라 결정되는 빔포밍 공분산 및 상기 방향 벡터에 따라 빔포밍 가중치를 생성하고, 상기 반향 제거된 입력결과들 및 상기 빔포밍 가중치에 기초하여 출력결과들을 제공하는 빔포밍기를 포함하는 타겟신호 추출 시스템.
- 현재프레임에 상응하는 현재프레임 입력결과들, 현재프레임 과거 입력결과들에 기초하여 반향 제거된 입력결과들을 생성하는 반향 제거기;이전프레임에 상응하는 이전프레임 입력신호 공분산 및 현재프레임의 상기 반향 제거된 입력결과들에 기초하여 생성되는 현재프레임 입력신호 공분산을 생성하고, 현재프레임의 상기 반향 제거된 입력결과들 및 이전프레임 빔포밍 가중치에 기초하여 현재프레임 분산 추정값을 생성하고, 상기 이전프레임에 상응하는 이전프레임 노이즈 공분산 및 현재프레임 분산 추정값에 기초하여 현재프레임 노이즈 공분산을 생성하며, 상기 현재프레임 입력신호 공분산, 상기 현재프레임 노이즈 공분산 및 이전프레임 방향 벡터에 기초하여 현재프레임 방향 벡터를 생성하는 방향 벡터 예측기; 및이전프레임 빔포밍 가중치, 현재프레임의 상기 반향 제거된 입력결과들 및 이전프레임 분산에 따라 현재프레임 빔포밍 분산 추정값을 생성하고, 이전프레임 역 공분산, 현재프레임의 상기 반향 제거된 입력결과들 및 현재프레임 빔포밍 분산 추정값에 기초하여 현재프레임 빔포밍 역 공분산을 생성하고, 상기 현재프레임 빔포밍 역 공분산 및 상기 현재프레임 방향 벡터에 따라 현재프레임 빔포밍 가중치를 생성하고, 현재프레임의 상기 반향 제거된 입력결과들 및 현재프레임 빔포밍 가중치에 기초하여 현재프레임 출력결과들을 제공하는 빔포밍기를 포함하는 온라인 타겟신호 추출장치.
- 제17항에 있어서,상기 온라인 타겟신호 추출장치는,상기 방향 벡터 예측기 및 상기 빔포밍기에 미리 결정된 마스크를 추가하여 상기 현재프레임 출력결과들을 제공하는 것을 특징으로 하는 온라인 타겟신호 추출장치.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US17/921,074 US20230178089A1 (en) | 2020-05-18 | 2021-05-07 | Beamforming method using online likelihood maximization combined with steering vector estimation for robust speech recognition, and apparatus therefor |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR1020200058882A KR20210142268A (ko) | 2020-05-18 | 2020-05-18 | 강인음성인식을 위한 방향벡터 추정을 겸한 온라인 우도최대화를 이용한 빔포밍 방법 및 그 장치 |
KR10-2020-0058882 | 2020-05-18 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2021235750A1 true WO2021235750A1 (ko) | 2021-11-25 |
Family
ID=78708776
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/KR2021/005759 WO2021235750A1 (ko) | 2020-05-18 | 2021-05-07 | 강인음성인식을 위한 방향벡터 추정을 겸한 온라인 우도최대화를 이용한 빔포밍 방법 및 그 장치 |
Country Status (3)
Country | Link |
---|---|
US (1) | US20230178089A1 (ko) |
KR (1) | KR20210142268A (ko) |
WO (1) | WO2021235750A1 (ko) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20240009758A (ko) * | 2022-07-14 | 2024-01-23 | 서강대학교산학협력단 | 강인한 음성인식을 위한 타겟 마스크 및 독립성분분석 기반의 실시간 빔포밍 및 방향 벡터 추정 방법 |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090164212A1 (en) * | 2007-12-19 | 2009-06-25 | Qualcomm Incorporated | Systems, methods, and apparatus for multi-microphone based speech enhancement |
KR20190073852A (ko) * | 2017-12-19 | 2019-06-27 | 서강대학교산학협력단 | 우도 최대화를 이용한 빔포밍 방법 |
KR20190091061A (ko) * | 2018-01-26 | 2019-08-05 | 서강대학교산학협력단 | 재귀적 최소 제곱 기법을 이용한 온라인 cgmm에 기반한 방향 벡터 추정을 이용한 음원 방향 추정 방법 |
KR102076760B1 (ko) * | 2018-09-19 | 2020-02-12 | 한양대학교 산학협력단 | 다채널 마이크를 이용한 칼만필터 기반의 다채널 입출력 비선형 음향학적 반향 제거 방법 |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101133308B1 (ko) | 2011-02-14 | 2012-04-04 | 신두식 | 에코제거 기능을 갖는 마이크로폰 |
-
2020
- 2020-05-18 KR KR1020200058882A patent/KR20210142268A/ko not_active Application Discontinuation
-
2021
- 2021-05-07 US US17/921,074 patent/US20230178089A1/en active Pending
- 2021-05-07 WO PCT/KR2021/005759 patent/WO2021235750A1/ko active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090164212A1 (en) * | 2007-12-19 | 2009-06-25 | Qualcomm Incorporated | Systems, methods, and apparatus for multi-microphone based speech enhancement |
KR20190073852A (ko) * | 2017-12-19 | 2019-06-27 | 서강대학교산학협력단 | 우도 최대화를 이용한 빔포밍 방법 |
KR20190091061A (ko) * | 2018-01-26 | 2019-08-05 | 서강대학교산학협력단 | 재귀적 최소 제곱 기법을 이용한 온라인 cgmm에 기반한 방향 벡터 추정을 이용한 음원 방향 추정 방법 |
KR102076760B1 (ko) * | 2018-09-19 | 2020-02-12 | 한양대학교 산학협력단 | 다채널 마이크를 이용한 칼만필터 기반의 다채널 입출력 비선형 음향학적 반향 제거 방법 |
Non-Patent Citations (1)
Title |
---|
조지원 등. 가산 잡음 또는 반향 환경에 강인한 음성인식을 위한 은닉 마르코프 모델 기반 특징 향상 방법. 한국통신학회지(정보와 통신). vol. 33, no. 9, PP. 17-23, September 2016. non-official translation (JO, Ji-Won et al. Hidden Markov Model-Based Feature Enhancement Method for Robust Speech Recognition in Additive Noise or Echo Environments. Journal of the Korea Institute of Information and Communication Engineering(Information and Communications Magazine)). [Retrieved on 05 July 2021]. Retrieved from <https://scienceon.kisti.re.kr/main/mainForm.do>. * |
Also Published As
Publication number | Publication date |
---|---|
KR20210142268A (ko) | 2021-11-25 |
US20230178089A1 (en) | 2023-06-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2009157715A2 (en) | Codebook design method for multiple input multiple output system and method for using the codebook | |
WO2017222140A1 (ko) | Cnn 기반 인루프 필터를 포함하는 부호화 방법과 장치 및 복호화 방법과 장치 | |
WO2009131376A2 (en) | Multiple antenna communication system including adaptive updating and changing of codebooks | |
WO2016089009A1 (en) | Method and cloud server for managing device | |
WO2014137159A1 (en) | Method and apparatus for applying secondary transforms on enhancement-layer residuals | |
WO2012077993A2 (ko) | 도어락 시스템 | |
WO2015183051A1 (en) | Soft buffer processing method and apparatus | |
WO2015139232A1 (zh) | 一种应用的推荐方法、系统及服务器 | |
WO2011076035A1 (zh) | 一种实现多卡槽访问的方法和装置 | |
WO2021112335A1 (ko) | 생성적 적대 신경망 기반의 분류 시스템 및 방법 | |
WO2020122593A1 (en) | Electronic device for attenuating at least part of signal received by antenna and method for controlling communication signal | |
WO2021235750A1 (ko) | 강인음성인식을 위한 방향벡터 추정을 겸한 온라인 우도최대화를 이용한 빔포밍 방법 및 그 장치 | |
WO2020060206A1 (en) | Methods for audio processing, apparatus, electronic device and computer readable storage medium | |
WO2017155137A1 (ko) | 빔포밍 방법 및 이를 위한 장치 | |
WO2022158686A1 (ko) | 암호화된 정보에 기초하여, 인공지능 모델을 이용한 추론을 수행하는 전자 장치 및 그 동작 방법 | |
WO2017018593A1 (ko) | 단말간 통신 시스템에서 전송 전력을 결정하는 장치 및 방법 | |
WO2011002260A2 (en) | Rotating reference codebook that is used in a multiple-input multiple-output (mimo) communication system | |
WO2021242063A1 (ko) | 초음파 투과 장치 및 파동 제어 방법 | |
WO2011040707A2 (en) | Multiple-input multiple-output communication system using explicit feedback | |
WO2019132553A1 (en) | Sound output system and voice processing method | |
WO2022124860A1 (ko) | 기지국 안테나의 기울기 각도 최적화 방법 및 장치 | |
WO2018143653A1 (ko) | 터치 입력 및 지문 입력을 인식하는 장치 및 방법 | |
WO2011083900A1 (en) | Codebook design method for multiple-input multiple-output (mimo) communication system and method for using the codebook | |
WO2021100936A1 (en) | Power factor adjustment method and apparatus in a waveguide circuit and a transmission line circuit, and power generating transmission line system using the same | |
WO2017014483A1 (ko) | 기술적 파급효과 분석 방법 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 21807737 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 21807737 Country of ref document: EP Kind code of ref document: A1 |
|
32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 14.04.2023) |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 21807737 Country of ref document: EP Kind code of ref document: A1 |