Abstract:
PURPOSE: A microphone array apparatus having a hidden microphone array structure and an acoustic signal processing apparatus including the same are provided to efficiently receive an acoustic signal by forming a microphone array including microphones. CONSTITUTION: A microphone array apparatus(100) includes a plurality of microphones(10,20,30,40). The microphones can be arranged in a linear or nonlinear structure. A cover(110) can have a hexahedron box shape. A unit(200) performing a specific function can be attached to a front side(111) of the cover. The interval between first type microphones covered with the cover can be narrowly arranged in comparison with the interval between second type microphones. The second type microphone is arranged in the inner groove region of a supporting plate(120).
Abstract:
본 발명은 히스토그램 정규화된 스태틱 특징 벡터 또는 히스토그램 정규화되지 않은 스태틱(static) 특징 벡터로부터 델타(delta) 특정 벡터를 구하여 히스토그램 정규화한 후 정규화된 델타 특징 벡터로부터 액셀레이션(acceleration) 특징 벡터를 구하여 히스토그램 정규화하는 방법 및 그 장치에 관한 것이다. 본 발명은 음성 신호에 대한 히스토그램 정규화된 스태틱 특징 벡터로부터 델타 특징 벡터를 구하여 히스토그램 정규화하는 단계 및 상기 히스토그램 정규화된 델타 특징 벡터로부터 액셀레이션 특징 벡터를 구하여 상기 액셀레이션 특징 벡터를 히스토그램 정규화하는 단계를 포함하는 것을 특징으로 하는 음성 특징 벡터의 정규화 방법을 제공한다. 본 발명에 의하면, 델타 히스토그램 정규화를 이용함으로써 학습에 이용된 깨끗한 음성 특징 벡터의 멀티 포인트 통계량까지 정규화가 가능한 이점이 있다. 히스토그램, 특징 벡터, static, delta, acceleration
Abstract:
PURPOSE: A gain controlling device and method thereof are provided to adjust bandwidth of a sound signal by using the updated weight. CONSTITUTION: A micro phone array includes at least two microphone which is arranged on the same surface. A frequency converting unit(110) converts a plurality of sound signal received from the microphone array into a signal of each frequency. A weight calculating unit(120) maintains a plurality sound signal which is converted into the frequency area. The weight calculating unit calculates a plurality of sound signal according to the frequency component. A scaling unit(130) adjusts an amplitude of a plurality of acoustic signals by using the calculated weight. The weight calculating unit calculates the weight.
Abstract:
본 발명은 음성 특징 벡터 변환 방법 및 장치에 관한 것으로, 음성 신호 중에서 음성 인식에 필요한 특징 벡터를 추출하고, 추출된 특징 벡터를 자동 연상 신경망을 이용하여 변환함으로써, 음성 인식 과정에서 잡음이 포함된 음성 특징 벡터가 들어오더라도, 강인한 특징 출력 값들을 얻을 수 있는 효과가 있다. 음성, 인식, 특징 벡터, 변환, 자동 연상 신경망
Abstract:
A method and an apparatus for correcting a sound source signal are provided to prevent a distortion of a sound source signal due to characteristic mismatching between individual microphones by correcting the sound source signal according to the calculated reference probability distribution. A probability distribution calculator calculates the probability distributions expressing the number of the sound source signals existing in each interval according to the size interval of the sound source signals. A probability distribution calculator includes an accumulator. The accumulator calculates and accumulates the probability distributions about the size of the sound source signals in each interval. A reference probability distribution calculator(211) calculates the reference probability distribution representing the probability distributions based on the calculated probability distributions. A signal corrector(212) corrects the sound source signals according to the calculated reference probability distribution. The reference probability distribution calculator calculates the reference probability distribution based on the accumulated probability distributions.
Abstract:
본 발명은 혼합 사운드로부터 잡음을 제거하는 방법 및 장치에 관한 것으로, 본 발명에 따른 잡음 제거 방법은 목표 사운드와 잡음을 포함한 음원 신호들을 입력받고, 입력된 음원 신호들로부터 음원 신호들 간의 속성 차이를 나타내는 하나 이상의 특징 벡터를 추출하고, 추출된 특징 벡터에 기초하여 음원 신호들에 대한 잡음 비율을 고려한 감쇄 계수를 산출하며, 산출된 감쇄 계수에 따라 음원 신호들로부터 생성된 출력 신호의 강도를 조절함으로써, 음향 센서를 통해 입력된 혼합 사운드로부터 잡음 신호를 제거하여 선명한 목표 음원 신호를 얻을 수 있다.
Abstract:
A method and an apparatus for extracting a target sound signal from a mixed signal are provided to extract a target audio signal in which PESQ(Perceptual Evaluation of Speech Quality) is high from the mixed signal by using an adaptive non-linear filter to an interference noise ratio. A suppressing signal beam former(222) produces a signal in which directivity is suppressed toward a target sound source direction. An emphasizing signal beam forming emphasizes sound pressure with regard to a specific target sound source. A microphone array improves an amplitude by giving proper weight to each signal. A beam former spatially reduces noise of an interference noise signal and a target signal. An adder adds a sound signal inputted through the microphone array. A delay-and-sum algorithm finds out a location of the sound source from a relatively delay time for which a signal reaches to the microphone.
Abstract:
A multi-stage speech recognition apparatus and method are provided to re-score a plurality of candidate words obtained in a primary recognition stage by using a temporal posterior feature vector to improve recognition performance. A multi-stage speech recognition apparatus includes a primary speech recognition unit(110) and a secondary speech recognition unit(130). The primary speech recognition unit performs primary speech recognition on a feature vector extracted from an inputted speech signal to generate a plurality of candidate words. The secondary speech recognition unit re-scores the plurality of candidate words by using a temporal posterior feature vector extracted from the speech signal to output a final recognition result.
Abstract:
A semiconductor package and a method for preparing the semiconductor package are provided to improve the confidence of package of a semiconductor by suppressing expansion of a packaging material with low modulus by using a packaging material with high modulus. A semiconductor package(100) comprises a semiconductor chip(110); a substrate(120) where the semiconductor chip is adhered; a wire(130) which connects electrically the semiconductor chip and the substrate; an external contact terminal which connects electrically the semiconductor chip and the outside; and a packaging material(136) which packages the wire and its surroundings and comprises a plurality of insulating materials(132, 134) of different physical properties. Preferably the plurality of insulating materials has different modulus.
Abstract:
본 발명은 혼합 사운드로부터 목표 음원 신호를 추출하는 방법 및 장치에 관한 것으로, 본 발명에 따른 목표 음원 신호 추출 방법은 마이크로폰 어레이를 통해 혼합 신호를 입력받고, 혼합 신호에 대하여 목표 음원 방향으로 지향성이 강조된 제 1 신호와 목표 음원 방향으로 지향성이 억제된 제 2 신호를 생성하며, 제 1 신호 및 제 2 신호 간의 비율에 기초하여 제 1 신호에 포함된 간섭 음원 신호를 마스킹하여 제 1 신호로부터 목표 음원 신호를 추출함으로써, 마이크로폰 어레이를 통해 입력된 복수 개의 사운드가 포함된 혼합 사운드로부터 특정 음원 신호를 선명하게 분리할 수 있다.