Abstract: A music tone pitch shift apparatus which converts an original audio signal into digital data by way of pulse code modulation (PCM), shifting the pitch, and converting the pitch shifted digital data into an analog signal. The PCM digital data is stored in a ring memory at a given sampling speed, and is read out of the memory by a pair of identical read circuits at a common read addressing speed corresponding to the desired pitch. One of the read circuits starts reading from the opposite address location to the other on the ring memory. Since the read addressing speed is set faster than the write addressing speed when increasing the pitch, and vice versa, overtaking or lapping between the addresses could occur.

Abstract: In a method of and an apparatus for the segmentation of speech, an acoustic speech signal is converted into N signals S.sub.i (f), each signal pertaining to a time interval i of N successive time intervals (1.ltoreq.i.ltoreq.N). For successive time intervals i a function c.sub.ij (FIG. 7) is then derived which is a measure of the agreement between the signals in the time intervals i and j. The middle m.sub.i of the pertaining function C.sub.ij is determined for each time interval i and those values i are determined which at least approximately correspond to zero-axis crossings with the same sign in a function d.sub.i (see FIG. 3b), d.sub.i indicating the difference between m.sub.i and i. The signal segmented in this way can be used for the derivation of diphones so that a library of diphones can be built up (see FIG. 1), or it can be used for the recognition of the speech utterance (see FIG. 2).

Abstract: There is provided an automatic voice recognition system which utilizes time encoded speech. Through the determination of zero crossing information and waveform parameters of an input voice signal, a stream of time encoded speech symbols is obtained. The stream of time encoded speech symbols is then converted into a matrix format for comparison with reference matrices formatted from time encoded symbols of reference words thereby to provide an output signal indicative of the content of the input voice signal.

Abstract: The process for varying the speed of a speech signal that involves splitting at least a portion of the speech frequency bandwidth into N narrow sub-bands, processing each sub-hand signal contents to derive therefrom magnitude data M(i, n) and phase data P(i, n), i=1, . . . , N being the subband index and n the time index. The M (i, n) sequence is converted into a sequence M'(n) by either duplicating one sample every K samples (K being an integer value derived from the desired slowing-down/speeding up ratio). The phase sequence P (i, n) is processed to derive therefrom an increment sequence D(i, n)=P(i, n)-P(i, n-1), which increment sequence is first converted into a D'(i, n) sequence by either dropping or duplicating one sample every K, samples, before being converted into P'(i, n)=P'(i, n)+D'(i, n). The P'(i, n), D'(i, n) sequences are converted back into sub-band signals contents, then combined together into the slowed-down/speeded-up speech signal.

Abstract: A pattern recognizer uses the method of n-tuples applied to a matrix containing (e.g.) features derived from input speech. A template store records during a training sequence the frequencies of occurrence of combinations of bits with each n-tuple group. During recognition frequencies associated with combinations which occur are used to form a measure of similarity.

Abstract: Syllables are recognized in voice data obtained from an input voice signal. Character arrays are used to represent the types and time frames of syllables with which the voice data are compared.

Abstract: A method and apparatus for generating, and storing for later retrieval, selected voice messages in response to selected verbal input trigger phrases. The voice signals including predetermined trigger phrases are transmitted from an input device such as a multiline telephone to a voice recognition system. The voice recognition system associates the trigger phrases with predetermined blocks of text message and in response to a sequence of trigger phrases, creates a report. The text is then output to a voice synthesizer system for the creation of synthetic voice signals corresponding to the full text of the report and the synthetic voice signals are output to a voice storage and forward system for later retrieval. The system also provides a printed copy of the full text of the report.

Abstract: A speech recognition apparatus has a speech input unit for inputting a speech; a speech analysis unit for analyzing the inputted speech to output the time series of a feature vector; a candidates selection unit for inputting the time series of a feature vector from the speech analysis unit to select a plurality of candidates of recognition result from the speech categories; and a discrimination processing unit for discriminating the selected candidates to obtain a final recognition result. The discrimination processing unit includes three components in the form of a pair generation unit for generating all of the two combinations of the n-number of candidates selected by said candidate selection unit, a pair discrimination unit for discriminating which of the candidates of the combinations is more certain for each of all .sub.n C.sub.

Abstract: A masking field network F.sub.2, is characterized through systematic computer simulations serves or a content addressable memory. Masking field network F.sub.2 receives input patterns from an adaptive filter F.sub.1 .fwdarw.F.sub.2 that is activated by a prior processing level F.sub.1. The network F.sub.2 activates compressed recognition close that are predictive with respect to the activation patterns flickering across F.sub.1, and competitively inhibits, or masks, codes which are unpredictive with respect to the F.sub.1 patterns. The masking field can simultaneously detect multiple groupings within its input patterns and assign activation weights to the recognition codes for these groupings which are predictive with respect to the contextual information embedded within the patterns and the prior learning of the network. Automatic rescaling of sensitivity of the masking field as the overall size of an input pattern changes, allows stronger activation of a code for the whole F.sub.

Abstract: An electronic voice recognition system utilizes a microphone having noise-cancelling characteristics and a standardizer network to achieve further noise reduction and signal standardization over a range of input levels. The output of the standardizer is coupled to voice processor circuitry which provides a computer-compatable output corresponding to the voice information received. The system may be embodied in a multi-station environment in which the voice reception stations are coupled to a remote central processing location. A variety of processing techniques, including analog and digital, parallel and sequential, may be utilized to pass the voice signals to the processors. Upon successful processing, a confirmation signal is passed to the appropriate voice-originating station.

Abstract: A method for forming a reference voice pattern to be stored in a library for use in identifying an unknown input sound.

Abstract: A continuous speech recognition system includes an automatic phonological rules generator which determines variations in the pronunciation of phonemes based on the context in which they occur. This phonological rules generator associates sequences of labels derived from vocalizations of a training text with respective phonemes inferred from the training text. These sequences are then annotated with their pheneme context from the training text and clustered into groups representing similar pronunciations of each phoneme. A decision tree is generated using the context information of the sequences to predict the clusters to which the sequences belong. The training data is processed by the decision tree to divide the sequences into leaf-groups representing similar pronunciations of each phoneme. The sequences in each leaf-group are clustered into sub-groups representing respectively different pronunciations of their corresponding phoneme in a give context. A Markov model is generated for each sub-group.

Abstract: A speech recognition system measures the values of at least two classes of features of an utterance: (1) a first class whose value is related to the frequency spectrum of the utterance, and (2) a second class whose value is related to the variation with time of the "first class" value of the utterance. Word baseforms are constructed from Markov model baseform units. Each output-producing transition of a baseform unit produces outputs from both classes. However, for each output-producing transition, the probabilities of producing outputs from the first class are independent of the probabilities of producing outputs from the second class.

Abstract: Improved speech sounds are produced from an all-software speech generating program on a computer which does not have a digital-to-analog converter by polling a first timer connected to the computer's speaker until a time-out is found, then applying a sample time value to a second timer so connected, and restarting the first timer. This results in the generation of a variable-duty-cycle carrier without interfering with the computer's operation. In an alternative embodiment, the first timer is used to generate an interrupt at regular intervals, and a sample time value from a storage buffer is then applied to the second timer.

Abstract: A continuous speech recognition unit using forward probabilities for recognizing continuous speech associated with standard patterns for given units of recognition comprises a standard template memory for storing Markov model standard templates of standard speech, which are composed of state sequences and transition probabilities between the states; an observation probability computing device for computing a forward probability for a feature vector time sequence; and a cumulative value computing device for determining a cumulative value based on the sum of previous cumulative values. The unit further comprises a matching pass memory for storing maximum values produced by the cumulative value computing means and a result processor for determining recognition results indicative of recognized words. The unit stores the transition giving the best probability in memory for each state and traces back the recognition result for the word sequence based on the transitions in memory.

Abstract: A phoneme estimator (12) in a speech-recognition system (10) includes trigger circuitry (18, 22) for identifying the segments of speech that should be analyzed for phoneme content. Speech-element processors (24, 26, and 28) calculate the likelihoods that currently received speech contains individual phonemes, but they operate only when the trigger circuitry identifies such segments. The computation-intensive processing for determining phoneme likelihoods is thus performed on only a small subset of the received speech segments. The accuracy of the speech-element processors (24, 26, and 28) is enhanced because these processors operate by recognition of patterns not only in elements of the data-reduced representations of the received speech but also in higher-ordered products of those elements; that is, these circuits employ non-linear modeling for phoneme identification.

Abstract: A method for creating word models for a large vocabulary, natural language dictation system. A user with limited typing skills can create documents with little or no advance training of word models. As the user is dictating, the user speaks a word which may or may not already be in the active vocabulary. The system displays a list of the words in the active vocabulary which best match the spoken word. By keyboard or voice command, the user may choose the correct word from the list or may choose to edit a similar word if the correct word is not on the list. Alternately, the user may type or speak the initial letters of the word. Then the recognition algorithm is called again satisfying the initial letters, and the choices displayed again. A word list is then also displayed from a large backup vocabulary. The best words to display from the backup vocabulary are chosen using a statistical language model and optionally word models derived from a phonemic dictionary.

Abstract: An interrogating device which includes a rule memory for storing forward inference rules utilizable in a forward inference control system or backward inference rules utilizable in a backward inference control system, a working memory for storing facts, and a forward inference control section for adding or deleting a fact relative to the working memory according to the forward inference rules or a backward inference control section for conducting a search according to the backwad inference rules.

Abstract: A system and method for tracking a number of subjects in a plurality of areas is shown to include a plurality of transmitters, wherein at least one transmitter is associated with each subject, each transmitter transmits a light based signal representative of an identifying code unique to that transmitter, a plurality of receivers, wherein at least one receiver is associated with each area, each receiver converts transmitted light based signals to electrical signals and validates the electrical signal to determine whether such electrical signals are representative of the unique identifying codes associated with the transmitters, and a central processing member, connected to each of the receivers, for recording those electrical signals which are representative of the unique identifying codes for recording the receiver which determined that such electrical signals are representative of the unique identifying codes associated with the transmitters and for determining in which area the transmitter is actually loca

Abstract: Voice feature data extracted from an input voice signal are stored in a first memory. Of the stored voice feature data, feature data having a predetermined duration defined by the output from a word boundary detection section are read out by a re-sampling section, and are stored in a second memory. The voice feature data which are normalized along the time base in this manner are supplied to a similarity computing section together with reference pattern data, and a category pattern corresponding to the resultant maximum similarity is determined in a determining section and outputted as a recognition result of the input voice.