Abstract: A system and method for noise reduction applied to a speech recognition front-end. An output of a front-end is optimized by giving, as a weight to the output for each band, a confidence index representing the remarkableness of the harmonic structure of observation speech. In a first method, when clean speech is estimated by executing MMSE estimation on a model that gives a probability distribution of noise-removed speech generated from observation speech, the posterior probability of the MMSE estimation is weighted using the confidence index as a weight. In a second method, linear interpolation is executed, for each band, between an observed value of observation speech and an estimated value of clean speech, with the confidence index serving as a weight. The first method and the second method can be combined.

Abstract: There is provided a method of producing a semiconductor wafer, including: forming a compound semiconductor layer on a base wafer by epitaxial growth; cleansing a surface of the compound semiconductor layer by means of a cleansing agent containing a selenium compound; and forming an insulating layer on the surface of the compound semiconductor layer. Examples of the selenium compound include a selenium oxide. Examples of the selenium oxide include H2SeO3. The cleansing agent may further contain one or more substances selected from the group consisting of water, ammonium, and ethanol. When the surface of the compound semiconductor layer is made of InxGa1-xAs (0?x?1), the insulating layer is preferably made of Al2O3, and Al2O3 is preferably formed by ALD.

Abstract: A probability model represented as the product of the probability distribution of a mismatch vector g (or clean speech x) with an observed value y as a factor and the probability distribution of a mismatch vector g (or clean speech x) with a confidence index ? for each band as a factor, executes MMSE estimation on the probability model, and estimates a clean speech estimated value x^. As a result, each band influences the result of MMSE estimation, with a degree of contribution in accordance with the level of its confidence. Further, the higher the S/N ratio of observation speech, the more the output value becomes shifted to the observed value. As a result, the output of a front-end is optimized.

Abstract: A voice activity detection method in a low SNR environment. The voice activity detection is performed by extracting a long-term spectrum variation component and a harmonic structure as feature vectors from a speech signal and increasing difference in feature vectors between speech and non-speech (i) using the long-term spectrum variation component feature or (ii) using a long-term spectrum variation component extraction and a harmonic structure feature extraction. A correct rate and an accuracy rate of the voice activity detection is improved over conventional methods by using a long-term spectrum variation component having a window length over an average phoneme duration of an utterance in the speech signal. The voice activity detection system and method provides speech processing, automatic speech recognition, and speech output capable of very accurate voice activity detection.

Abstract: A semiconductor wafer comprises, on a semiconductor crystal layer forming wafer, a first semiconductor crystal layer, a second semiconductor crystal layer, and a third semiconductor crystal layer in this order, wherein both the etching rates of the first semiconductor crystal layer and the third semiconductor crystal layer by a first etching agent are higher than the etching rate of the second semiconductor crystal layer by the first etching agent, and both the etching rates of the first semiconductor crystal layer and the third semiconductor crystal layer by a second etching agent are lower than the etching rate of the second semiconductor crystal layer by the second etching agent.

Abstract: A mirror surface angle adjusting device includes: a pivot plate (10) which supports a mirror (6) and is attached to a holder (9) pivotably three-dimensionally on a pivot shaft (76); a bracket (8) disposed on a vehicle body; a mirror tilting drive section (11) contained between the bracket (8) and the holder (9) attached to the bracket (8); an operation shaft (44A, 44B) of the mirror tilting drive section (11), the operation shaft penetrating through the holder (9) and being connected to the pivot plate (10); a detection rod (101A, 101B) which moves following the pivot plate (10); an angle detection section (100) which detects a tilt angle of the pivot plate (10); a pivot depressed portion (80) into which the pivot shaft can be press-fitted; and a gear positioning rib (69) provided on an inner wall of the holder (9) to position the mirror tilting drive section (42).

Abstract: There is provided a compound semiconductor wafer that is suitably used to form a plurality of different types of devices such as an HBT and an FET thereon. The semiconductor wafer includes a first semiconductor, a carrier-trapping layer that is formed on the first semiconductor and has an electron-trapping center or a hole-trapping center, a second semiconductor that is epitaxially grown on the carrier-trapping layer and serves as a channel in which a free electron or a free hole moves, and a third semiconductor including a stack represented by n-type semiconductor/p-type semiconductor/n-type semiconductor or represented by p-type semiconductor/n-type semiconductor/p-type semiconductor, where the stack is epitaxially grown on the second semiconductor.

Abstract: A speech feature extraction apparatus, speech feature extraction method, and speech feature extraction program. A speech feature extraction apparatus includes: first difference calculation module to: (i) receive, as an input, a spectrum of a speech signal segmented into frames for each frequency bin; and (ii) calculate a delta spectrum for each of the frame, where the delta spectrum is a difference of the spectrum within continuous frames for the frequency bin; and first normalization module to normalize the delta spectrum of the frame for the frequency bin by dividing the delta spectrum by a function of an average spectrum; where the average spectrum is an average of spectra through all frames that are overall speech for the frequency bin; and where an output of the first normalization module is defined as a first delta feature.

Abstract: A method and system for generating training data for a target domain using speech data of a source domain. The training data generation method including: reading out a Gaussian mixture model (GMM) of a target domain trained with a clean speech data set of the target domain; mapping, by referring to the GMM of the target domain, a set of source domain speech data received as an input to the set of target domain speech data on a basis of a channel characteristic of the target domain speech data; and adding a noise of the target domain to the mapped set of source domain speech data to output a set of pseudo target domain speech data.

Abstract: A semiconductor wafer includes a first semiconductor, and a second semiconductor formed directly or indirectly on the first semiconductor. The second semiconductor contains a first impurity atom exhibiting p-type or n-type conductivity, and a second impurity atom selected such that the Fermi level of the second semiconductor containing both the first and second impurity atoms is closer to the Fermi level of the second semiconductor containing neither the first impurity atom nor the second impurity atom, than the Fermi level of the second semiconductor containing the first impurity atom is. For example, the majority carrier of the second semiconductor is an electron, and the Fermi level of the second semiconductor containing the first and second impurity atoms is lower than the Fermi level of the second semiconductor containing the first impurity atom.

Abstract: The present invention relates to a system, method and program for speech recognition. In an embodiment of the invention a method for processing a speech signal consists of receiving a power spectrum of a speech signal and generating a log power spectrum signal of the power spectrum. The method further consists of performing discrete cosine transformation on the log power spectrum signal and cutting off cepstrum upper and lower terms of the discrete cosine transformed signal. The method further consists of performing inverse discrete cosine transformation on the signal from which the cepstrum upper and lower terms are cut off. The method further consists of converting the inverse discrete cosine transformed signal so as to bring the signal back to a power spectrum domain and filtering the power spectrum of the speech signal by using, as a filter, the signal which is brought back to the power spectrum domain.

Abstract: There is provided a decorated sheet having a good adhesiveness with a molding resin regardless of a pattern of a colored part. A decorated sheet comprising: a transparent layer coating on the entire surface of a thermoplastic substrate sheet or at least in the areas of said substrate sheet having no coloring layer formed, said transparent layer coated by a transparent ink with as common varnish composition as the ink for coloring layer formation, and an adhesive layer formed as the outermost layer of said decorated sheet on the side provided with the colored layers of the substrate sheet. A molded article is fabricated by integrating the decorated sheet with a molding resin.

Abstract: An apparatus, program product and method is provided for separating a target voice from a plurality of other voices having different directions of arrival. The method comprises the steps of disposing a first and a second voice input device at a predetermined distance from one another and upon receipt of voice signals at said devices calculating discrete Fourier transforms for the signals and calculating a CSP (cross-power spectrum phase) coefficient by superpositioning multiple frequency-bin components based on correlation of the two spectra signals received and then calculating a weighted CSP coefficient from said two discrete Fourier-transformed speech signals. A target voice is separated when received by said devices from other voice signals in a spectrum by using the calculated weighted CSP coefficient.

Abstract: The present invention relates to a method, preprocessor, speech recognition system, and program product for extracting a target speech by removing noise. In an embodiment of the invention target speech is extracted from two input speeches, which are obtained through at least two speech input devices installed in different places in a space, applies a spectrum subtraction process by using a noise power spectrum (U?) estimated by one or both of the two speech input devices (X?(T)) and an arbitrary subtraction constant (?) to obtain a resultant subtracted power spectrum (Y?(T)). The invention further applies a gain control based on the two speech input devices to the resultant subtracted power spectrum to obtain a gain-controlled power spectrum (D?(T)). The invention further applies a flooring process to said resultant gain-controlled power spectrum on the basis of arbitrary Flooring factor (?) to obtain a power spectrum for speech recognition (Z?(T)).

Abstract: When a waveform monitor apparatus 1 of an injection molding machine M for monitoring an operation waveform at least during molding is configured by mounting the waveform monitor apparatus 1 on the injection molding machine M for performing molding according to a specific molding system, a parting opening detecting means Fp for detecting change data of a parting opening Lm to a time during molding, and an operation waveform displaying means Fd for displaying at least change data from a resin filling start is into a mold 2 to a cooling time termination to of the mold 2, detected by the parting opening detecting means Fp, on a waveform displaying portion 5 of a screen 4v on a display 4 attached to a molding machine controller 3 are provided.

Abstract: There is provided a method of producing a semiconductor wafer, including: forming a compound semiconductor layer on a base wafer by epitaxial growth; cleansing a surface of the compound semiconductor layer by means of a cleansing agent containing a selenium compound; and forming an insulating layer on the surface of the compound semiconductor layer. Examples of the selenium compound include a selenium oxide. Examples of the selenium oxide include H2SeO3. The cleansing agent may further contain one or more substances selected from the group consisting of water, ammonium, and ethanol. When the surface of the compound semiconductor layer is made of InxGa1-xAs (0?x?1), the insulating layer is preferably made of Al2O3, and Al2O3 is preferably formed by ALD.

Abstract: A speech signal processing system which outputs a speech feature, divides an input speech signal into frames so that each pair of consecutive frames have a frame shift length equal to at least one period of the speech signal and have an overlap equal to at least a predetermined length, applies discrete Fourier transform to each of the frames, calculates a CSP coefficient for the pair, searches a predetermined search range in which a speech wave lags a period equal to at least one period to obtain the maximum value of the CSP coefficient for the pair, and generates time-series data of the maximum CSP coefficient values arranged in the order in which the frames appear. A method and a computer readable article of manufacture for the implementing the same are also provided.

Abstract: A system and method for noise reduction applied to a speech recognition front-end. An output of a front-end is optimized by giving, as a weight to the output for each band, a confidence index representing the remarkableness of the harmonic structure of observation speech. In a first method, when clean speech is estimated by executing MMSE estimation on a model that gives a probability distribution of noise-removed speech generated from observation speech, the posterior probability of the MMSE estimation is weighted using the confidence index as a weight. In a second method, linear interpolation is executed, for each band, between an observed value of observation speech and an estimated value of clean speech, with the confidence index serving as a weight. The first method and the second method can be combined.

Abstract: A speech feature extraction apparatus, speech feature extraction method, and speech feature extraction program. A speech feature extraction apparatus includes: first difference calculation module to: (i) receive, as an input, a spectrum of a speech signal segmented into frames for each frequency bin; and (ii) calculate a delta spectrum for each of the frame, where the delta spectrum is a difference of the spectrum within continuous frames for the frequency bin; and first normalization module to normalize the delta spectrum of the frame for the frequency bin by dividing the delta spectrum by a function of an average spectrum; where the average spectrum is an average of spectra through all frames that are overall speech for the frequency bin; and where an output of the first normalization module is defined as a first delta feature.

Abstract: It is an objective of the present invention to form a favorable interface between an oxide layer and a group 3-5 compound semiconductor using a practical and simple method. Provided is a semiconductor wafer comprising a first semiconductor layer that is a group 3-5 compound not containing arsenic and that lattice matches or pseudo-lattice matches with InP; and a second semiconductor layer that is formed to contact the first semiconductor layer, is a group 3-5 compound semiconductor layer that lattice matches or pseudo-lattice matches with InP, and can be selectively oxidized relative to the first semiconductor layer.