Abstract: An inputted sound signal is sampled at intervals over a period and cepstrum coefficients are calculated from the sampled values. Cepstrum sum, distance and/or power are calculated and compared with appropriately preselected threshold values to distinguish voice (vowel) intervals and noise intervals. The ratio of the length of the voice intervals to the sampling period is considered to determine whether the sampled inputted sound signal represents voice or noise.

Abstract: A digital signal processor implementation of dynamic time warping for automatic speech recognition using a single memory array that stores only one sequence of accumulated correspondence values and detecting whether or not the constraint on horizontal path compression has been exceeded by storing negatives of calculated accumulated correspondence values to indicate horizontal path movement. The accumulated correspondence values are stored in the array in descending order; and memory locations of that array are reused as accumulated correspondence values are calculated representing the correspondence between a plurality of reference time frames and a plurality of unknown time frames representing an unknown word or utterance. When path movement is from an adjacent horizontal correspondence node to the present node, the negative of the calculated accumulated correspondence is written back into a memory location associated with the present node.

Abstract: A method and arrangement for a speech recognition system employs channel bank information to represent speech. The method considers background noise included with the speech. The method includes determining three energy levels for each channel the first representative of background noise energy, the second representative of the input frame energy and the third representative of the word template frame energy. Values representing energy level differentials are assigned at each channel. If the second energy level is less than the first energy level, then a predetermined constant value is assigned at the particular channel. These values are combined to generate a distance measure depicting the similarity between the two frames.

Abstract: A speech coding system includes apparatus for generating a variable threshold dependent upon the power of an input speech signal, and a comparator for comparing the power of the input speech signal with the variable threshold value to generate a discriminating signal for discriminating between a period when a speech continues and a period when the speech pauses, to change the coding operation for the input speech signal in accordance with the level of the discriminating signal, thereby forming voiced and unvoiced frames independently of each other.

Abstract: A text editor is connected to a speech recognizing unit for editing preferably spoken input text using a display speech. For each text word (including digits), and each punctuation mark that can be recognized and is contained in a dictionary, a token is stored for holding information on character count, capitalization, left and right concatenation of the respective item, and for providing fields for context conditions. For each segment or entity recognized spoken text, a respective character string and associated token is transferred to storage in the editor to allow automatic formatting and correct displaying or printing of the text, including spaces and capitalization where required. Tokens are updated during editing to reflect modifications such as in the beginning of a sentence or in concatenation. Switching to spelling mode is provided for entering single spelled characters in case where a word cannot be recognized or where spelling is desired.

Abstract: A vocoder for matching an input speech signal with a reference speech signal on the basis of mutual angular data developed through spherical coordinate conversion of a plurality of formant frequencies obtained from the input and reference speech signals.

Abstract: A logarithmic frequency spectrum related to an input signal is converted by the use of an inverse Fourier transform into a cepstrum. The cepstrum has a first and a second frequency component which have a first peak and a second peak spaced apart from the first peak by a preselected period on an axis of frequency, respectively. The second frequency component is processed into a peak controlled frequency component having a controlled peak coincident with the first peak. The peak controlled frequency component and the first frequency component are summed up to produce an ultimate frequency component which corresponds to the value of the envelope parameter.

Abstract: An apparatus for detecting the modulation rate of voice-band data signals. The detector determines the level occupancy L.sub.p, the partial entropy E.sub.p, and the zero-crossing count N.sub.zc for the input signal and compares these values with threshold values to determine the data rate.

Abstract: A method and system for cancelling noise from sources that are distributed over a region, whereby two sensors are located so that a first sensor will detect both voice signals and noise signals, and a second sensor will detect only the noise signals. The voice signals picked up at the second sensor are negligible, and the noise signals picked up at both sensors are correlated. The signals output from each sensor are connected to a predetermined number of narrowband filters in order to divide each respective signal into a predetermined number of frequencies, such as 15 for example. Thereafter, both signals are combined to cancel effectively the noise component from the signal output having both voice and noise to leave a voice signal that is substantially noise free.

Abstract: A speech encoding process, wherein a first sequence of input data representative of a written version of a message to be coded is encoded to provide a first encoded speech sequence corresponding to the written version of the message to be coded, and a second sequence of input data derived from speech defining a spoken version of the same message is analyzed by a linear predictive codeing analyzer and encoding circuit to provide a second encoded speech sequence corresponding to the spoken version of the message to be coded. The codes of the corresponding written message and the codes of the spoken message are then combined in a control circuit encompassing an adaptation algorithm, and a composite encoded speech sequence is generated corresponding to the message from the combination of the first encoded speech sequence of the written version of the message and encoded intonation parameters of speech included in a portion of the second encoded speech sequence corresponding to the spoken version of the message.

Abstract: An speech analysis and synthesis system where pitch information for excitation is transmitted during voiced segments of speech and modified residual information for excitation is transmitted during unvoiced speech segments along with linear predictive coded (LPC) parameters. The speech analysis portion of the system uses a pitch detection circuit to determine when the speech is voiced or unvoiced and to calculate the pitch information during voiced segments. A multi-pulse excitation forming circuit generates the modified residual signal which is obtained from the cross correlation of the residual signal and the LPC-recreated original signal. The pitch detection circuit controls a multiplexer which selects either the output of the multi-pulse excitation forming circuit or the output of the pitch detection circuit for transmission as the excitation information with LPC parameters to the synthesizer portion of the system.

Abstract: Control reference pattern candidates corresponding to a verification reference patterns of a registered speaker are synthesized by connecting unit speech patterns of a plurality of speakers. A plurality of control reference patterns similar to the verification reference pattern are determined from among the control reference pattern candidates. First dissimilarity between an input pattern of a speaker to be verified and the verification reference pattern specified by the registered speaker and second dissimilarity between the input pattern and the control reference patterns specified by the registered speaker are calculated. The speaker to be verified is judged as the registered speaker on the basis of the first and second dissimilarities.

Abstract: An audio signal recording and reproducing apparatus digitizes sound into an electric audio signal, data compresses the signal into compressed data to reduce information quantity, records the compressed data into a semiconductor memory, and reproduces an audio signal from the recorded data. The apparatus includes a plurality of input compression and output extension conversion tables which are used for compression conversion of input signal and extension conversion of output signal, respectively. Input compression and output extension conversion table selection counters are count-controlled by difference data representing the change of the audio signal. The apparatus selects one among a plurality of input compression and output extension conversion tables in accordance with count values of the selection counters. The conversion tables are arranged in sequence and contain successively larger ranges of values.

Abstract: Apparatus for encoding speech using a code excited linear predictive (CELP) encoder using a virtual searching technique during speech transitions such as from unvoiced to voiced regions of speech. The encoder compares candidate excitation vectors stored in a codebook with a target excitation vector representing a frame of speech to determine the candidate vector that best matches the target vector by repeating a first portion of each candidate vector into a second portion of each candidate vector. For increased performance, a stochastically excited linear predictive (SELP) encoder is used in series with the adaptive CELP encoder. The SELP encoder is responsive to the difference between the target vector and the best matched candidate vector to search its own overlapping codebook in a recursive manner to determine a candidate vector that provides the best match. Both of the best matched candidate vectors are used in speech synthesis.

Abstract: Inputted voice signals are analyzed in units of syllables. Each syllable is compared with standard syllables preregistered in a memory and thereby recognized but corrections may be made on erroneous recognitions by referencing a dictionary or entering a command. Each standard pattern is associated with phonological information on the neighborhood in a voice signal from where it was extracted and may be updated by another pattern having the same phonological information. Temporal sequences of correct and erroneous recognitions of individual syllables as well as whole syllables are stored and referenced in determining whether a standard pattern should be updated. A maximum pattern number may be set for each syllable.

Abstract: A speech synthesis system comprises an input terminal for accepting text code, accent code, and phrase code.

Abstract: In speech recognition it is advantageous to take account of noise levels both in recognition and training. In both processes signals reaching a microphone are digitized and passed through a filter bank to be separated into frequency channels. In training, a noise estimator and a masker are used with a recognizer to prepare and store probability density functions (p.d.f.s) for each channel partially defining Markov models of words to be recognized. The p.d.f.s are derived only from input signals above noise levels but derivation is such that the whole of each p.d.f. is represented. In recognition, "distance" measurements on which recognition is based are derived for each channel. If the signal in a channel is above noise then the distance is determined, by the recognizer, from the negative logarithm of the p.d.f. but if a channel signal is below noise then the distance is determined from the negative logarithm of the cumulative distance of the p.d.f. to the noise level.

Abstract: A voice pattern in the form of a matrix and comprised of a plurality of frames, each including time-spectral information and temporal information, is formed from an unknown input voice signal. The voice pattern is compared with each of the voice patterns of a library of known voices partly to select a plurality of candidate voices. Each of the library voices has a predetermined eigenvector and an inner product frequency distribution of inner products between the eigenvector and the frames of its voice pattern. Then, inner products between the voice pattern of the input voice signal and the eigenvector of each of the candidate library voice are calculated. One of the plurality of candidate library voices whose predetermined inner product frequency distribution is most similar to one of the thus calculated inner product frequency distributions is selected to identify the input voice signal.

Abstract: Pattern matching, for speech recognition, wherein distances are determined between the input pattern and many reference patterns. However the distances are determined with the reference pattern at a time slant to the input pattern and the determination is repeated for different starting values of the input patterns. Dynamic programming (DP) matching is conducted based on these distances. Boundary conditions may be determined for different starting values independent of other starting values because allowed paths do not reverse directions on the scale of the gradations.

Abstract: A sub-band coder utilizes one or more bit allocation tables to dynamically distribute the channel bit capacity bandwidth among the frequency bands according to the desired output quality of speech rather than by means of complex algorithms or simulation techniques. Multiple bit assignment tables are provided to allow various quality levels to be traded off as increasing bit rate demands are placed upon the transmission system. The technique is used for a single coder to achieve a minimum bit rate for a desired given level of subjective quality in speech output or may be used in a shared bit resource to maintain equal and minimum quality degradation for all users. The quality tables determine the number of bits to be dropped from the encoded representation of each signal sample to minimize the transmission load for a given coder without sacrificing speech quality to an unacceptable degree.