Abstract: A method of analyzing data, particularly a speech signal, first pre-processes the signal by performing analog-to-digital conversion and cepstral analysis, producing a sequence of data frames. Then the sequence of data frames is partitioned into a plurality of data blocks. The data blocks may be subjected to further analysis, for example, by introducing them to a plurality of neural networks. The system may be implemented using either hardware or software or a combination thereof.

Abstract: A video signal is converted into compressed frame signals, each comprising a coefficient signal of block coefficient signals for each frame in a video image. Each of the block coefficient signals represents the pixels in a pixel map block (420) with the coefficients in a hybrid polynomial. The hybrid polynomial contains discrete cosine terms, a constant term separated from the discrete cosine terms, and polynomial terms extracted from the discrete cosine terms. Each block coefficient signal contains a background component representing a coefficient of the constant term, a linear component representing the coefficients of the polynomial terms, and a nonlinear component representing the coefficients of the discrete cosine terms.

Abstract: The invention is a process for producing N-substituted glucamines, a process for oxidizing an N-substituted glucamine with an oxidizing microbe or extract thereof, and a process for oxidizing an N-substituted glucamine with an oxidizing microbe or extract thereof and reducing the oxidized N-substituted glucamine to an N-substituted-1-deoxynojirimycin. In addition, a one pot process for producing N-substituted-1-deoxynojirimycin from D-glucose is disclosed.

Abstract: The invention is a process for producing N-substituted glucamines, a process for oxidizing an N-substituted glucamine with an oxidizing microbe or extract thereof, and a process for oxidizing an N-substituted glucamine with an oxidizing microbe or extract thereof and reducing the oxidized N-substituted glucamine to an N-substituted-1-deoxynojirimycin. In addition, a one pot process for producing N-substituted-1-deoxynojirimycin from D-glucose is disclosed.

Abstract: In a speech-recognition system having a plurality of classifiers, a voting window includes a sequence of outputs from each of the classifiers. At each interval in the voting window, the outputs are compared to determine a winning output. A spoken sound is identified by determining which classifier generates the greatest number of winning outputs in the voting window.

Abstract: Boundaries of spoken sounds in continuous speech are identified by classifying delimitative sounds to provide improved performance in a speech-recognition system. Delimitative sounds, those portions of continuous speech that occur between two spoken sounds, are recognized by the same method used to recognize spoken sounds. Recognition of delimitative sounds is accomplished by training a learning machine to act as a classifier which implements a discriminant function based on a polynomial expansion.

Abstract: A computing device, which may be implemented as an integrated circuit, is constructed of a microprocessor and one or more neural network co-processors. The microprocessor normally executes programs which transfer data to the neural network co-processors, which are used to compute complicated mathematical functions. Direct Memory Access (DMA) is also used to transfer data. Each neural network co-processor interfaces to the microprocessor in a manner substantially similar to that of a conventional memory device. The co-processor does not require any instructions and is configured to execute mathematical operations simply by being pre-loaded with gating functions and weight values. In addition, the co-processor executes a plurality of arithmetic operations in parallel, and the results of such operations are simply read from the co-processor.

Abstract: An artificial neuron, which may be implemented on an integrated circuit, has only one significant processing element in the form of an adder. Inputs are first fed through gating functions to produce gated inputs. These gated inputs are then added together to produce a sum which is added to a weight to produce the neuron output.

Abstract: A method is implemented in hardware or software type neural network, the neural network is constructed of neurons or neuron circuits each having only one significant processing element in the form of a multiplier. Each neuron applying a gating function to each of neural network inputs. The neural network utilizes a training algorithm which does not require repetitive training and which yields a global minimum to each given set of input vectors.

Abstract: A circuit and method for computing an exponential signal x.sup.g is provided. The circuit includes a logarithm converter which converts an input signal to binary word that represents the logarithm of an input signal x. A first shift register shifts the binary word in a bit-wise fashion to produce a first intermediate value; while a second shift register shifts the binary word in a bit-wise fashion to produce a second intermediate value. The shift registers may be implemented using multiplexers. The shifting operations are equivalent to multiplying the intermediate values by a factor which is a power of two. The first intermediate value is either added to or subtracted from the second intermediate value to produce a combined value. An inverse-logarithm converter converts the combined value to the exponential signal.

Abstract: A system polynomial is determined using a plurality of system I/O data, wherein the system polynomial expresses the system output in terms of the system input, and wherein the system polynomial has at least one linear term and at least one nonlinear term. A control polynomial is determined, the control polynomial having at least one cancellation term and at least one control term, the at least one cancellation term calculated to cancel the at least one nonlinear term of the system polynomial, and the at least one control term calculated to control the at least one linear term of the system polynomial. A control output signal is generated based on the control polynomial and the control input signal.

Abstract: An artificial neuron, which may be implemented on an integrated circuit, has only one significant processing element in the form of an adder. Inputs are first fed through gating functions to produce gated inputs. These gated inputs are then added together to produce a sum which is added to a weight to produce the neuron output.

Abstract: A method for storing data in a memory device is provided. The method stores blocks of data elements in a two-dimensional address space such that the blocks overlap in the address space. The data elements may be digital words which represent exponents of a polynomial expansion.

Abstract: A system polynomial is determined using a plurality of system I/O data, wherein the system polynomial expresses the system output in terms of the system input, and wherein the system polynomial has r' terms including at least one linear term and at least one nonlinear term, the r' terms found using a regression subsets technique. A control polynomial is determined, the control polynomial having at least one cancellation term and at least one control term, the at least one cancellation term calculated to cancel the at least one nonlinear term of the system polynomial, and the at least one control term calculated to control the at least one linear term of the system polynomial. A control output signal is generated based on the control polynomial and the control input signal.

Abstract: A neural network, which may be implemented either in hardware or software, is constructed of neurons or neuron circuits each having only one significant processing element in the form of a multiplier. A hidden neuron in the neural network generates an output based on the product of a plurality of functions. The neural network utilizes a training algorithm which does not require repetitive training and which yields a global minimum to each given set of input vectors.

Abstract: A speech-recognition system for recognizing isolated words includes pre-processing circuitry for performing analog-to-digital conversion and cepstral analysis, and a plurality of neural networks which compute discriminant functions based on polynomial expansions. The system may be implemented using either hardware or software or a combination thereof. The speech wave-form of a spoken word is analyzed and converted into a sequence of data frames. The sequence of frames is partitioned into data blocks, and the data blocks are then broadcast to a plurality of neural networks. Using the data blocks, the neural networks compute polynomial expansions. The output of the neural networks is used to determine the identity of the spoken word. The neural networks utilize a matrix-inversion or alternatively a least-squares estimation training algorithm which does not require repetitive training and which yields a global minimum to each given set of training examples.

Abstract: An artificial neuron, which may be implemented either in hardware or software, has only one significant processing element in the form of a multiplier. Inputs are first fed through gating functions to produce gated inputs. These gated inputs are then multiplied together to produce a product which is multiplied by a weight to produce the neuron output.

Abstract: A bond inspection technique which determines the integrity of a plurality of package leads (13) bonded to a plurality of contact areas (12) on a semiconductor chip (11). A bonding process heats each package lead (13) bonded to each contact area (12). A camera (16) forms an infra-red intensity image at a predetermined time of the semiconductor chip (11) and ports image data to a computer (18). Infra-red intensity radiated from each bond on the semiconductor chip (11) is compared by the computer (18) with infra-red intensity data of known good and bad bonds. The comparison of each bond determines bond integrity of the semiconductor chip (11).

Abstract: Light (19) is projected at an incidence angle (29) onto a plurality of leads (12, 13). The light (19) is simultaneously reflected from each of the plurality of leads (12, 13). The light that is simultaneously reflected (24, 26) from each lead (12, 13) is detected. A cotangent function of the incidence angle (29) is utilized to detect an amount of displacement (32) of at least one of the plurality of leads.

Abstract: A non-destructive method for testing quality of at least one bond which physically and electrically couples a package lead to an integrated circuit. A thermal gradient is created across the at least one bond which causes heat transfer thru the at least one bond. Heat transfer is measured which is proportional to area at a bond interface. The measured heat transfer thru the at least one bond is compared with heat transfer data taken under substantially equal conditions of known good bonds thereby determining quality of the at least one bond. The heat transfer is indirectly measured by creating a mechanical vibration at an input area and by measuring the time it takes the mechanical vibration to reach an output area after the mechanical vibration has traveled through the at least one bond.