Abstract: The encoder unit of the vocoder device includes the AR code-book, the MA code-book, and the voice source code-book storing code words each corresponding to a set of the AR, the MA, and the voice source parameters, respectively, which are obtained beforehand by means of the "analysis by synthesis" of a multitude of speech waveform examples and then clustering the resulting respective parameters. The AR preliminary selector, the MA preliminary selector, and the voice source preliminary selector select from respective code-books a predetermined finite number of code words approximating the input speech signal, and in synchronism with the voice source position detected by the voice source position detector the speech synthesizer synthesizes a number of synthesized speech waveforms corresponding to the combinations of the selected AR, MA, and voice source parameters.

Abstract: A compressed digital speech signal is encoded to provide a transmission error-resistant transmission signal. The compressed speech signal is derived from a digital speech signal by performing a pitch search on a block obtained by dividing the speech signal in time to provide pitch information for the block. The block of the speech signal is orthogonally transformed to provide spectral data, which is divided by frequency into plural bands in response to the pitch information. A voiced/unvoiced sound discrimination generates voiced/-unvoiced (V/UV) information indicating whether the spectral data in each of the plural bands represents a voiced or an unvoiced sound. The spectral data in the plural bands are interpolated to provide spectral amplitudes for a predetermined number of bands, independent of the pitch.

Abstract: A speech recognition method in which input speech signals are converted to digital signals and then time sequentially converted to cepstrum coefficients or logarithmic spectra. Dynamic spectrum time sequence is obtained by time frequency filtering of cepstrum coefficients, or masked spectrum time sequence is obtained by time frequency masking of the logarithmic vector time sequence. Based on the dynamic cepstrum time sequence or masked spectrum time sequence obtained in this manner, speech is recognized.

Abstract: Methods and apparatus for speech coding are disclosed for converting analog speech signals to digital speech signals for transmission. The speech coder, utilizing CELP techniques, includes a first filter for filtering out the spectral information from the speech signal. The spectral information is provided for transmission. A second filter is provided for filtering out the pitch information from the speech signal and such pitch information is also provided for transmission. A codevector generator determines, in one embodiment, the characteristics of a bi-pulse codevector representative of the speech signal. In this embodiment the impulse response of the first filter is truncated for determining the codevector characteristics. In this embodiment it is also preferred to determine the codevector characteristics by conducting a numerator only search in relation to a traditional fraction used for determining codevectors.

Abstract: A Forward Feed Neural Network is disclosed using data flow techniques on a data flow microprocessor. As a result of this invention, a neural network is provided that has the capacity of "learning" to distinguish among patterns of data which may differ recognizably from idealized cases, and is able to perform pattern recognition faster while utilizing less memory and fewer clock cycles than neural networks implemented on sequential processors. This implementation is simpler and faster because of an inherent similarity between the flow of information in the brain and in data flow architecture.

Abstract: Novel neural networks and novel methods for training those networks are disclosed. The novel networks are feedforward networks having at least three layers of neurons. The training methods are easy to implement, converge rapidly, and are guaranteed to converge to a solution. A novel network structure is used, in which each corner of the input vector hypercube may be considered separately. The problem of mapping may be reduced to a sum of corner classification sub-problems. Four efficient, alternative classification methods for use with the novel neural networks are also disclosed.