Abstract: A multi-pulse speech coding method and apparatus capable of encoding speech at a bit rate of 16 kbps or less. The method determines the location and amplitude of a pulse by searching through all of the samples of a criterion function, modifying all of the samples of the criterion function, and them repeating the pulse search. After the predetermined number of pulses have been determined, the method modifies the amplitude of the determined pulse, modifies the criterion function at the location where the pulses are set, and repeats such pulse amplitude modification. The method is, therefore, capable of modifying a pulse amplitude by using only a minimum amount of computation. As compared to the amount of computerization required by a method of the kind which modifies pulse amplitude in a pulse search loop.

Abstract: An automatic speech recognition system has a multi-mode training capability using a set of previously stored templates of a limited number of predetermined seed words to train the templates for a vocabulary of words. The training speech samples each includes a vocabulary word juxtaposed with a seed word. An averager module maintains an active average template for each of the word units of the training speech samples including the seed word units, and the active average templates are used to continuously update the seed template set as they are used in the training speech samples. The preferred training procedure employs training phrases each having a vocabulary word embedded between two seed words, and two seed template sets are used in succession, the first being composed of single-digit words, and the second composed of carrier words.

Abstract: A sound generator for storing and reproducing audio signal with reduced memory storage, which is accomplished by: (a) the technique of Adaptive Differential Pulse Code Modulation (ADPCM), and (B) the elimination of silence period from memory storage but regenerating the silence signal at output.

Abstract: A speech synthesizing apparatus includes an interpolation pitch calculation circuit wherein an interpolation pitch is calculated at every frame periods based upon a speech parameter. Specifically, a difference between a target value and a present value is stored in a shift register of 16 bits, and then, in order to omit the figures below the first place of decimals and to divide the difference by the number of subframes, the shift register is shifted rightward by 7 bits. If a numeral value of the shift register is negative, "1" is added to the number of units of the numeral value in an adding circuit such that an absolute value of the numeral value becomes smaller. When the numeral value of the shift register is positive, the numeral value outputted from the shift register is set in an interpolation pitch register as it is and, when negative, an output of the adding circuit is set in the interpolation pitch register.

Abstract: A cost-effective word recognizer. Each frame of spoken input is compared to a set of reference frames. The comparison is equivalent to embodying the reference frame as an LPC inverse filter, and is preferably done in the autocorrelation domain. To avoid the instability and computational difficulties which can be caused by a high-gain LPC inverse filter, a noise floor is introduced into each reference frame sample. Thus, for each input speech frame, a scalar measures its similarity to each of the vocabulary of reference frames.To achieve connected word recognition based on this similarity measurement, a dynamic programming algorithm is used in which time warping to match a sample to a reference is in effect permitted, and in which matching is performed with unconstrained endpoints.

Abstract: A voice language translator, suitable for implementation in hand-held size, is disclosed. The voice language translator includes: a key pad (20); a display system (17); a language cartridge(s) (45); a voice recognition module (49); a voice synthesizer (47); a speaker (39); a microphone (41); and a programmed CPU (43). Prior to use as a translator, the voice language translator is trained to the voice of a user. During training, a series of words and phrases to be spoken by the user are displayed, or spoken, in the language of the user. As the user speaks the words and phrases, the voice recognition circuit produces a digitally coded voice pattern that uniquely identifies the way in which the user spoke the words and phrases. The voice patterns produced by the voice recognition circuit are analyzed and stored, preferably in the cartridge.

Abstract: Power information from an input signal is used to detect the presence of speech. A reference point is established as the moment of detection of the speech. During a period between the reference point and a subsequent point distant from the reference point by a predetermined range, the input signal is linearly changed to a corresponding signal having a predetermined period. Feature parameters are extracted from the signal with the predetermined period. The feature parameters are replaced by preset noise parameters in a portion having no speech component therein. Standard speech patterns of particular preset words are determined and similarities between the extracted feature parameters containing the noise parameters and the standard patterns are calculated and mutually compared. The foregoing steps are performed while the separations are varied within the predetermined range. Similar steps are performed as the reference point is shifted by a unit period, and similarities are calculated and mutually compared.

Abstract: A speech analysis method which includes the steps of detecting a maximum-level position in that portion of an input speech signal which exists in a period equal to the pitch period of the input speech signal from a predetermined one of periodically-generated timing pulses, tracing the speech signal from the maximum-level position in a time reversing direction to find a zero-crossing point where the level of the traced signal is first reduced to zero, extracting a one-pitch signal which starts from the zero-crossing point and has the duration equal to the pitch period of the input speech signal, from the speech signal, and carrying out Fourier transform for the one-pitch signal to obtain a spectrum of the input speech signal.

Abstract: By reconciling differences between the estimator and the filter of a code excited linear predictive (CELP) voice coder, higher quality is achieved in the output speech. The pulse amplitudes and pitch tap gain are solved for simultaneously to minimize the estimator bias in the CELP excitation. Increased signal to noise ratio is accomplished by modifying the pitch predictor such that the pitch synthesis filter accurately reflects the estimation procedure used to find the pitch tap gain, and by improving the excitation analysis technique such that the pitch predictor tap gain and codeword gain are solved for simultaneously, rather than sequentially. These modifications do not result in an increased transmission rate or significant increase in complexity of the CELP coding algorithm.

Abstract: A text-to-speech conversion system converts specified text strings into corresponding strings of consonant and vowel phonemes. A parameter generator converts the phonemes into formant parameters, and a formant synthesizer uses the formant parameters to generate a synthetic speech waveform. A library of vowel allophones are stored, each stored vowel allophone being represented by formant parameters for four formants. The vowel allophone library includes a context index for associating each said vowel allophone with one or more pairs of phonemes preceding and following the corresponding vowel phoneme in a phoneme string. When synthesizing speech, a vowel allophone generator uses the vowel allophone library to provide formant parameters representative of a specified vowel phoneme. The vowel allophone generator coacts with the context index to select the proper vowel allophone, as determined by the phonemes preceding and following the specified vowel phoneme.

Abstract: An unknown voiceband digital data modem signal is classified as being generated by one of a plurality of possible digital data modem signal sources, e.g., CCITT V.29, CCITT V.32, CCITT V.33 or the like digital data modems. Classification is achieved by employing a blind, i.e., self-recovering, adaptive equalizer to remove effects of linear channel impairments and to generate a sequence of magnitude estimates at the symbol rate of the unknown voiceband digital data modem signal. The sequence of magnitude estimates is compared to predetermined representations of known possible voiceband digital data modem signals and the results of the comparison are used to identify the digital data modem signal source of the signal. In one example, the predetermined representations are templates of conditional density functions of magnitude estimates obtained from known voiceband digital data modem signals generated by corresponding digital data modem signal sources.

Abstract: A method of digitally evaluating the frequency and the phase of signals in the form of digitized samples, the method being wherein it comprises in succession:a stage during which numbers corresponding to the samples of the signals to be analyzed are processed in order to convert them into the form of an analytic signal whose real portion coincides with said signal to be analyzed; and, in parallel therewith:a stage of estimating the parameters to be analyzed which is performed in an overall manner on the basis of estimators and of selection criteria by working on the phase of the signal without using any operator of the Fourrier transform type of any hypothesis test either separately or simultaneously; anda stage of estimating the differences between the real signal as taken in this way and the signal obtained from the estimated parameters, thereby making it possible to deliver data in digital form relating to the quality of the analyzed signal and to the reliability of the estimated values.

Abstract: In coded speech communication, discrete speech samples are analyzed to generate a first signal indicating the fine pitch structure of the speech samples and a second signal indicating their spectral characteristic. The amplitudes and locations of main excitation pulses are determined from the fine pitch structure and spectral characteristic and a third signal indicating the determined pulse amplitudes and locations is generated. The difference between the speech samples and the main excitation pulses is detected and used in auxiliary excitation pulse calculation to determine gain and index values of auxiliary excitation pulses by retrieving stored auxiliary excitation pulses from a code book so that the retrieved auxiliary excitation pulses approximate the difference.

Abstract: A speech coder employs vector quantization of LPC parameters, interpolation, and trellis coding for improved speech coding at low bit rates (400 bps). The speech coder has an LPC analysis module for converting input speech to LPC parameters, an LSP conversion module for converting LPC parameters into line spectrum frequencies (LSP) data, and a vector quantization and interpolation (VQ/I) module for encoding the LSP data into vector indexes for transmission by applying LPC spectral amplitude as weighting coefficients to the LSP data. The VQ/I module outputs one vector index for every two LPC frames in order to reduce the transmission bit rate, and the omitted frames are interpolated on the receiving end. A decoder correspondingly decodes incoming indexes to LPC parameters and synthesizes them into output speech. Trellis coders with an adaptive tracking function encode the pitch and gain parameters of the LPC frames. A universal codebook stores codewords according to a plurality of accents.

Abstract: For hard of hearing people, speech sounds are recognized electronically and displayed on eyeglasses at syllable speed, consonants are shown as a symbol resembling a hand sign, and vowels by the symbol's location in a quadrant display, or by color.

Abstract: An electronic hand-held, talking learning aid is disclosed. The learning aid includes a MOS speech synthesizer chip having an active surface area on the order of 45,000 square mils. The disclosed speech synthesizer chip includes a digital lattice filter, a voiced/unvoiced excitation circuit, a speech parameter interpolator, an input parameter decoder, a digital-to-analog converter and associated timing circuits. The learning aid is also provided with a microprocessor which functions as a controller for controlling the operation of the unit. A small speaker is driven by the digital-to-analog converter on the speech synthesis chip and a keyboard and display device are strobed by the microprocessor controller. Features include modes in which a speech synthesizer recites instructions or questions to the operator who must properly respond.

Abstract: An apparatus for drilling the pronunciation of language has a transducer for forming drill voice signals representing a practice utterance of a trainee and a reproducer for generating reference voice signals representing model pronunciation of the utterance. The drill and reference voice signals are compared by a processor and a sensible signal is produced to indicate whether the practice utterance corresponds to the model pronunciation.

Abstract: A method of encoding speech sounds to facilitate their transmission to and reconstruction at a remote receiver. A transmitter and a receiver have identical filters and identical codebooks containing prestored excitation vectors which model quantized speech sound vectors. The speech sound vectors are compared with filtered versions of the codebook vectors. The filtered vector closest to each speech sound vector is selected. During the comparison, filtration parameters derived by backward predictive analysis of a series of previously selected filtered codebook vectors are applied to the filter. The transmitter sends the receiver an index representative of the location of the selected vector within the codebook. The receiver uses the index to recover the selected vector from its codebook, and passes the recovered vector through its filter to yield an output signal which reproduces the original speech sound sample.

Abstract: In a pulse searching arrangement (65) of a multi-pulse voice encoder comprising a linear prediction residual signal producing arrangement (61), a pitch period is predicted in a cross-correlation domain in each subframe of a frame of an input voice signal. For this purpose, a local signal producing circuit comprises a first cross-correlator (66) for producing a first cross-correlation signal related to a linear prediction residual signal. A buffer memory (32) produces an output cross-correlation signal from subframe to subframe. Supplied with a sum signal and controlled by a pitch period signal representative of pitch periods of the input voice signal, a pitch synthesizing filter (71) produces a synthesized signal, which is added to provide the sum signal to a signal representative of excitation pulses. Responsive to the synthesized signal, a second cross-correlator (67) produces a second cross-correlation signal, which is subtracted from the output cross-correlation signal to provide the local signal.

Abstract: A voice recognition system for selecting word templates necessary for voice recognition from among a plurality of word templates. The system includes an input phoneme extraction unit for extracting distinctive phonemes which can be clearly recognized from an input pattern of the voice, and all phonemes possibly existing in the input pattern, and a dictionary phoneme extraction unit for extracting the distinctive phonemes which can be clearly recognized and can be previously determined at each word template, and all phonemes possibly existing at each word template. A pre-selector selects valid word templates for performing voice recognition based on phoneme information extracted by the input phoneme extraction unit and the dictionary phoneme extraction unit.