Abstract: An apparatus for generating a set of acoustic prototype signals for encoding speech includes a memory for storing a training script model comprising a series of word-segment models. Each word-segment model comprises a series of elementary models. An acoustic measure is provided for measuring the value of at least one feature of an utterance of the training script during each of a series of time intervals to produce a series of feature vector signals representing the feature values of the utterance. An acoustic matcher is provided for estimating at least one path through the training script model which would produce the entire series of measured feature vector signals. From the estimated path, the elementary model in the training script model which would produce each feature vector signal is estimated. The apparatus further comprises a cluster processor for clustering the feature vector signals into a plurality of clusters.

Abstract: A SOUND RECOGNITION PROCESS comprises training a sound recognition system to a shape Hidden Markov Model (HMM) from a shape codebook of shape codewords, and to a separate gain HMM from a gain codebook of gain codewords. An unknown token is assigned to the sound category associated with the shape HMM and gain HMM with the greatest probability product. Shape and gain codewords replacements for incoming Linear Predictive Coded (LPC) vectors are selected by choosing the codeword with minimum distortion between the codeword and a combination of the vector and some of the preceding codewords, exponentially weighted by time.

Abstract: A digital voice processing system wherein voice processing functions are run in software. This application of software allows a modular structure because the application software resides in boards that are coupled to a host computer. With this structure, the software can be updated as required and the capacity of the system can be readily expanded to meet increased needs.

Abstract: A speech coding apparatus includes multipliers and prediction filters which successively process a plurality of signal vectors obtained from an index 2.sup.M and dimension N code book to obtain a reproduced speech signal. Error detectors are provided which find the error between the input speech signal and reproduced speech signal. Evaluators are also provided which calculate the optimum signal vectors giving the smallest errors. The multipliers are connected to a reduced code book, which is constituted of n number of code book blocks of index 2.sup.M/n and dimension N/n (where n is an integer of two or more). There are n number of multipliers, n number of prediction filters, n number of error detectors, and n number of evaluators corresponding to the code book blocks.

Abstract: An apparatus for encoding voice spectrum envelop parameters forms a phoneme matrix by combining a certain number of phoneme vectors, and effects matrix quantization by using this phoneme matrix as a unit. The apparatus performs restricted time-direction deformation of an input phoneme matrix, such as by shifting, compression, or expansion in time-direction, to output a finite number of deformed phoneme matrices. The input phoneme matrix is formed by combining, in time-direction, a certain number of phoneme vectors composed of spectrum parameters representing information on the spectrum of an input voice signal. A code book is used for storing a second number of phoneme matrix code words which are compared with the deformed phoneme matrices provided by restricted time-direction deformation. The distances between the deformed phoneme matrices of the input phoneme matrix and the phoneme matrix code words, which are successively read out from the code book, are calculated.

Abstract: A language generator for a speech recognition apparatus scores a word-series hypothesis by combining individual scores for each word in the hypothesis. The hypothesis score for a single word comprises a combination of the estimated conditional probability of occurrence of a first class of words comprising the word being scored, given the occurrence of a context comprising the words in the word-series hypothesis other than the word being scored, and the estimated conditional probability of occurrence of the word being scored given the occurrence of the first class of words, and given the occurrence of the context. An apparatus and method are provided for classifying multiple series of words for the purpose of obtaining useful hypothesis scores in the language generator and speech recognition apparatus.

Abstract: A speech encoder uses a soft interpolation decision for spectral parameters. For each frame, the encoder first calculates the residual energy for interpolated spectral parameters, and then calculates the residual energy for non-interpolated spectral parameters. The encoder then compares these residual energy calculations. If the encoder determines that the interpolated spectral parameters yields the lowest residual energy, it indicates to a far-end decoder to use the interpolated values for the current frame. Otherwise, it indicates to the far-end decoder to use the non-interpolated values for the current frame. The encoder signals the far-end decoder as to which spectral parameters (interpolated or non-interpolated values) to use by encoding and transmitting a special signalling bit.

Abstract: A new method for Code Excited Linear Predictive (CELP) coding of speech reduces the computational complexity by removing a convolution operation from a recursive loop used to poll the adaptive code book vectors. In a preferred embodiment, an impulse function of a short term perceptually weighted filter is first convolved with perceptual weighted target speech and the result cross-correlated with each vector in the codebook to produce an error function. The vector having the minimum error function is chosen to represent the particular speech frame being examined.

Abstract: A signal processor, which receives autocorrelation and cross-correlation coefficients of a system under study, provides predictor and filter coefficients of the system in an optimal manner, and allows the use of any number of available parallel processing units. The signal processor may be implemented in a fully parallel or fully sequential manner, or in a "parallel-partitioned" implementation.

Abstract: A gain-shape vector quantization apparatus is provided for encoding and decoding, to transmit and receive compressed speech signals. A selected plurality of vectors are read from a code book based upon an index signal. The vectors are added in an adder and synthesis filtered by a synthesis filter, in either order, to produce an output. This output is subtracted from an input speech signal to produce an error signal. An evaluation unit produces an index to select the plurality of vectors read from the code book memory based on the error signal in order to minimize this error signal. The evaluation unit produces gain adjusting signals which can be used to adjust gains of the vectors read from the code book. In an encoder, signals indicative of the gain adjusting signal and the index signal are transmitted by a transmitter of the encoder to send a quantized speech signal to a receiver of a decoder.

Abstract: A method and apparatus for finding the best or near best binary classification of a set of observed events, according to a predictor feature X so as to minimize the uncertainty in the value of a category feature Y. Each feature has three or more possible values. First, the predictor feature value and the category feature value of each event is measured. The events are then split, arbitrarily, into two sets of predictor feature values. From the two sets of predictor feature values, an optimum pair of sets of category feature values is found having the lowest uncertainty in the value of the predictor feature. From the two optimum sets of category feature values, an optimum pair of sets is found having the lowest uncertainty in the value of the category feature. An event is then classified according to whether its predictor feature value is a member of a set of optimal predictor feature values.

Abstract: A code excited linear prediction (CELP) type speech signal coding system is provided, a code vector obtained by applying linear prediction to a vector of a residual speech signal of white noise is stored in a code book. A pitch prediction vector obtained by applying linear prediction to a residual signal of a preceding frame is given a delay corresponding to a pitch frequency and added to the code vector. Use is made of an impulse vector obtained by applying linear prediction to a residual signal vector of impulses having a predetermined relationship with the vectors of the white noise code book. Variable gains are given to at least the above code vector and impulse vector, a reproduced signal is produced, and this reproduced signal is used for identification of the input speech signal. Thus, a pulse series corresponding to the sound source of voiced speech sounds is created.

Abstract: Efficient coding speech information for low rate (e.g., 600 bps) channels using a four frame superframe (SF) includes: (1) coding spectral information using alternative quantizers one of which is chosen for each superframe so that 3 bits/SF identify the optimal quantizer and 28-32 bits/SF contain the quantized spectral information; (2) coding pitch using 5 bits/SF if voiced and if unvoiced assigning the pitch bits to error correction; (3) coding energy using 9-12 bits/SF by a 4d vector quantizer (4dvQ); and (4) coding voicing using 3-4 bits/SF by a 4d VQ, for a total of 54 bits/SF including 1 sync bit and 0-1 error correction bits. When combined with a unique perceptual weighting scheme, output speech quality comparable to that of vocoders operating at almost four times the channel capacity is obtained.

Abstract: An inner product computing unit computes inner products of an input pattern whose category is unknown, and orthogonalized dictionary sets of a plurality of reference patterns whose categories are known. A nonlinear converting unit nonlinearly converts the inner products in accordance with a positive-negative symmetrical nonlinear function. A neural network unit or a statistical discriminant function computing unit performs predetermined computations of the nonlinearly converted values on the basis of preset coefficients in units of categories using a neural network or a statistical discriminant function. A determining section compares values calculated in units of categories using the preset coefficients with each other to discriminate a category to which the input pattern belongs.

Abstract: A data compression system in which a dictionary stored strings of characters and an encoder matches the longest of the stored string with a current string of a data stream input to the encoder. The index of the longest matched stored string is output by the encoder and the dictionary is updated by a new string consisting of the previous match concatenated with the first two characters only of the present match. If the present match has only one or two characters, it is added without reduction.

Abstract: A priority assignment method and device are set forth for assigning a priority to a selected speech frame coded by a linear predictive coder based on at least two of: an energy of the speech frame, a log spectral distance between a frame and a frame immediately previous, and a pitch predictor coefficient for the selected speech frame. The invention protects against loss of perceptually important and hard-to-reconstruct speech frames.

Abstract: An optimal parametric signal processor, which receives autocorrelation coefficients, provides lattice coefficients in an optimal manner, and allows the use of any number of available parallel processing units. The signal processor may be implemented in a fully parallel or fully sequential manner, or in a "parallel-partitioned" implementation which provides the benefits of parallel processing, manageable hardware complexity, and optimal signal processing for a given number of available processors. By using a "lattice basic cell" structure, hardware complexity is minimized.

Abstract: A speech-recognition system includes a speech-recognition arrangement (1) and a learning device (5) for adapting the speech-recognition arrangement to the characteristic pronunciation of a given speaker during an adaptive learning mode. Certain words are applied during a learning procedure to the speech-recognition arrangement via a keyboard and, in addition, the sound signals associated with these certain words and originating from the given speaker are fed-in via a microphone. The sound signals corresponding to a spoken text are fed into the speech-recognition arrangement via a storage medium which can be scanned by a reproducing device. The reproducing device may be implemented in the form of a recording and reproducing device (7). The sound signals which are fed in to the system during a learning procedure of the speech-recognition arrangement are first recorded onto the storage medium and immediately thereafter are scanned from the storage medium and are fed into the speech-recognition arrangement.

Abstract: An electronic device comprises a first memory for storing a plurality of words which are to be displayed on a display coupled to the first memory and a second memory for storing a plurality of voice information codes each corresponding to a portion of a different one of the plurality of words stored in the first memory. The second memory also stores a specific code for distinguishing each of the plurality of voice information codes and stores each of the plurality of voice information codes and the specific code so that the voice information codes and the specific code are alternatively read out.

Abstract: Features are extracted from a sampled input signal by performing first linear predictive analyses of different first orders p on the sample values and performing second linear predictive analyses of different second orders q on the residuals of the first analyses. An optimum first order p is selected using information entropy values representing the information content of the residuals of the second linear predictive analyses. One or more optimum second orders q are selected on the basis of changes in these information entropy values. The optimum first and second orders are output as features. Further linear predictive analyses can be carried out to obtain higher-order features. Useful features are obtained even for nonstationary input signals.