Abstract: A method of breaking up a vocal signal into binary frames of a predetermined duration. The frames are grouped together in packets of successive frames by associating a predictive filter with each frame of a packet. Furthermore, the coefficients of each predictive filter are quantified by taking into account the stable or non-stable configuration of the vocal signal.

Abstract: As a step in finding the one most likely word sequence in a spoken language system, an N-best search is conducted to find the N most likely sentence hypotheses. During the search, word theories are distinguished based only on the one previous word. At each state within a word, the total probability is calculated for each of a few previous words. At the end of each word, the probability score is recorded for each previous word theory, together with the name of the previous word. At the end of the sentence, a recursive traceback is performed to derive the list of the N best sentences.

Abstract: A 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.

Abstract: The generation of multipulse excitation codes by digitizing an original speech, partitioning the digitized signal into a number of samples, pre-emphasizing the samples, producing linear predictive reflection coefficients from said samples, quantizing these reflection coefficients, converting the quantized reflection coefficients to spectral coefficients and subjecting the spectral coefficients to pitch analysis to obtain a spectral residual signal.

Abstract: In an adaptive subband excited transform speech encoding system, a range of quantizers are available and are dynamically selected for each window of speech. The quantizers designated for individual subbands are determined to minimize mean squared error distortion in the recreated signal while using no more than a predetermined number of quantization bits per window of speech.

Abstract: Pitch periods for a long term predictor included in a speech codec are searched in two searching stages. In the first searching stage, probable pitch periods are searched skipping a constant number of pitch periods, and in the second searching stage, pitch periods including the pitch period determined in the first searching stage and pitch periods neighboring the pitch period on both sides are searched.

Abstract: A speech recognition system in which speech information inputted from a microphone is recognized and a speech signal corresponding to the speech information is generated and converted into a set of characters equivalent to keystroke information normally inputted from a keyboard, the set of characters being supplied to a keyboard buffer, so that an application program running on a computer system executes a desired job.

Abstract: A method and a system for synthesizing speech from unrestricted text, based on the principle of associating a written string of text with a sequence of speech features vectors that most probably model the corresponding speech utterance. The synthesizer is based on the interaction between two different Ergodic Hidden Markov Models: an acoustic model reflecting the constraints on the acoustic arrangement of speech, and a phonetic model interfacing phonemic transcription to the speech features representation.

Abstract: This invention provides a novel speech coding system which recursively executes a filter-applied "Toeplitz characteristic" by causing a drive signal (i.e., an excitation signal) to be converted into a "Toeplitz matrix" when detecting a pitch period in which distortion of the input vector and the vector subsequent to the application of filter-applied computation to the drive signal vector in the pitch forecast called either "closed loop" or "compatible code book" is minimized. The vector quantization method substantially making up the speech coding system of the invention is characteristically used by the system.

Abstract: In a known method of recognizing a word string in a speech signal, a new specific organization of the storage locations in the memory containing the trace-back addresses is proposed. Furthermore, a step for generating the information for these storage locations is proposed which makes it possible to determine not only the word string with the best similarity, but also a specific number of further word strings of decreasing similarity. Thus, the additional computing capacity required for determining these further word strings is small relative to the computing capacity required for the remainder of the recognition process.

Abstract: The pitch estimation method is improved. Sub-integer resolution pitch values are estimated in making the initial pitch estimate; the sub-integer pitch values are preferably estimated by interpolating intermediate variables between integer values. Pitch regions are used to reduce the amount of computation required in making the initial pitch estimate. Pitch-dependent resolution is used in making the initial pitch estimate, with higher resolution being used for smaller values of pitch. The accuracy of the voiced/unvoiced decision is improved by making the decision dependent on the energy of the current segment relative to the energy of recent prior segments; if the relative energy is low, the current segment favors an unvoiced decision; if high, it favors a voiced decision.

Abstract: A method and apparatus are provided for identifying one or more boundaries of a speech pattern within an input utterance. One or more anchor patterns are defined, and an input utterance is received. An anchor section of the input utterance is identified as corresponding to at least one of the anchor patterns. A boundary of the speech pattern is defined based upon the anchor section. Also provided are a method and apparatus for identifying a speech pattern within an input utterance. One or more segment patterns are defined, and an input utterance is received. Portions of the input utterance which correspond to the segment patterns are identified. One or more of the segments of the input utterance are defined responsive to the identified portions.

Abstract: In a digital computer, there is provided a method of recognizing speech, comprising the steps of: entering a cohesive speech segment; determining gross acoustic attributes of the entered segment; determining fine acoustic attributes of the entered segment; assigning at least one subsyllable to the entered segment based on the gross and fine acoustic attributes determined; repeating the foregoing steps on successive cohesive speech segments to generate at least one sequence of subsyllables; converting the sequence of subsyllables into a sequence of syllables by finding the sequence of subsyllables in a table in which predetermined subsyllable sequences correspond with respective syllables and syllable sequences; combining the converted sequence of syllables into words; and verifying the conformance of the words to a first predetermined set of grammatical rules. An apparatus implementing the method is also disclosed.

Abstract: A speech system recognizes words from a spoken phrase that conform to checksum constraints. Grammar rules are applied to hypothesize words according to the checksum constraints. The checksum associated with the phrase is thus inherent in the grammar. Sentences which do not meet a predetermined checksum constraint are not valid under the grammar rules and are therefore inherently rejected. The checksum constraints result in increased recognition accuracy.

Abstract: A method of inputting Chinese characters into a computer directly from Mandarin speech which recognizes a series of monosyllables by separately recognizing syllables and Mandarin tones and assembling the recognized parts to recognize the mono-syllable using Hidden Markov Models. The recognized mono-syllable is used by a Markov Chinese Language Model in a Linguistic decoder section to determine the corresponding Chinese character A Mandarin dictation machine which uses the above method, using a speech input device to receive the Mandarin speech and digitizing it so a personal computer can further process that information. A pitch frequency detector, a Voice signal pre-processing unit, a Hidden Markov Model processor, and a training facility are all attached to the personal computer to perform their associated functions of the method above.

Abstract: A document edition system includes a microphone to input audio information; an analog to digital converter to convert the audio information into digital information; a memory unit to store the digital audio information into an external storage unit as information to be added to document information; a sound pressure display control unit to display the sound pressure information of the digital audio information to a display device; a digital to analog converter to regenerate the digital audio information from the memory unit into speech; and a device to add the audio information to an arbitrary position of the document information. The sound pressure display control unit can display the speech regenerating position from the digital to analog converter onto a sound pressure display waveform and also includes a device for displaying the sound pressure information by a predetermined pattern width or a tone.

Abstract: The pitch estimation method is improved. Sub-integer resolution pitch values are estimated in making the initial pitch estimate; the sub-integer pitch values are preferably estimated by interpolating intermediate variables between integer values. Pitch regions are used to reduce the amount of computation required in making the initial pitch estimate. Pitch-dependent resolution is used in making the initial pitch estimate, with higher resolution being used for smaller values of pitch. The accuracy of the voiced/unvoiced decision is improved by making the decision dependent on the energy of the current segment relative to the energy of recent prior segments; if the relative energy is low, the current segment favors an unvoiced decision; if high, it favors a voiced decision.

Abstract: A sound synthesizer which may be associated with a personal computer and including apparatus for employing the output of a noise generator which is cataloged to provide a multiplicity of waveforms and apparatus for receiving the multiplicity of waveforms and creating therefrom desired sound signals, thus providing a synthesized sound output.

Abstract: For speech signals that represent a wave form and include a sequence of samples, a method is provided for modifying the time scale of the speech signals. The method includes estimating the number of the samples that constitutes a pitch period. The estimate is made by a plurality of pitch estimators that operate in parallel and process peak and/or valley measurements of the wave form. The method also includes combining a first group of samples with a second group of samples to form a combined group. Each group consists of the same number of sequential samples from the sequence of samples. The number of samples in each group is equal to the estimated number of samples that constitutes a pitch period. The second group of samples immediately follows the first group in the sequence of samples. According to another feature of the invention, the method includes determininq whether the time scale of the speech signals is to be compressed or expanded.

Abstract: In a digital computer, a method for speech recognition includes steps of sampling a speaker's speech and providing speech data sample segments of predetermined length at predetermined sampling intervals based on changes in energy in the speech. Cohesive speech segments, which correspond to intervals of stable vocoids, changing vocoids, frication, and silence, are identified from the speech data sample segments, and are assigned frames of subsyllables. Each cohesive segment corresponds to at least one respective frame, and each frame includes at least one of a plurality of subsyllables that characterizes predetermined gross and fine phonetic attributes of the respective cohesive segment. The subsyllables are located in a first lookup table mapping sequences of subsyllables into syllables, and the syllables are combined into words by locating words in another lookup table. The conformance of sequences of the words to a set of predetermined checking rules is checked, and a recognition result is reported.