Abstract: A device for sound synthesis intended to generate a desired acoustic signal includes a first signal source which emits a periodic signal having a given repetition frequency as a representation of the voiced parts of the desired acoustic signal, a second signal source which emits an aperiodic signal or a noise signal as a representation of the unvoiced parts of the desired sound signal, a combination circuit which combines the signals of the two signal sources with each other, and a filter circuit having a variable transmission function for processing the combined signal into the desired output signal. A third signal source emits a modulated noise signal consisting of a train or sequence of noise bursts of comparatively short duration, whose temporal envelope is synchronous with the temporal envelope of the periodic signal and which invariably have at least approximately the same energy.

Abstract: A continuous speech prefiltering system for use in continuous speech recognition computer systems. The speech to be recognized is converted from utterances to frame data sets, which frame data sets are smoothed to generate a smooth frame model over a predetermined number of frames. A resident vocabulary is stored within the computer as clusters of word models which are acoustically similar over a succession of frame periods. A cluster score is generated by the system, which score includes the likelihood of the smooth frames evaluated using a probability model for the cluster against which the smooth frame model is being compared. Cluster sets having cluster scores below a predetermined acoustic threshold are removed from further consideration. The remaining cluster sets are unpacked for determination of a word score for each unpacked word.

Abstract: A speech categorization system includes first and second timers which generate first and second measured durations indicative of duration of selected higher and lower amplitude segments included in a voice message. A higher amplitude segment is classified in a first category when the first and second measured durations corresponding to the higher amplitude segment and an adjacent lower amplitude segment satisfy a classification test, and a counter counts the number of the higher amplitude segments classified in the first category. To detect spoken itemized words such as numbers or letters, the classification test can include the requirement that the lower amplitude segment be relatively short as compared to the preceding higher amplitude segment.

Abstract: In distinguishing between voiced and unvoiced speech elements use is made of the fact that the spectra of voiced sounds lie predominantly at or below about 1 kHz, and the spectra of unvoiced sounds lie predominantly at or above about 2 kHz. A change from a voiced sound to an unvoiced sound or vice versa always produces a clear shift of the spectrum, and that without such a change, there is no such clear shift. From the lower- and higher-frequency energy components, a measure of the location of the spectral centroid is derived which is used for a first decision. Based on the difference between two successive measures, a second decision is made by which the first can be corrected.

Abstract: A novel spectral interpolation and efficient excitation codebook search method developed for a Code-Excited Linear Predictive (CELP) speech coder is set forth. The interpolation is performed on an impulse response of the spectral synthesis filter. As the result of using this new set of interpolation parameters, the computations associated with an excitation codebook search in a CELP coder are considerably reduced. Furthermore, a coder utilizing this new interpolation approach provides noticeable improvement in speech quality coded at low bit-rates.

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 for positioning excitation pulses for a linear predictive coder (LPC) operating according to the multi-pulse principle, i.e. a number of such pulses are positioned at specific time points and with specific amplitudes. The time points and the amplitudes are determined from the predictive parameters (a.sub.k) and the predictive residue signal (d.sub.k), by correlation between a speech representative signal (y) and a composed synthesized signal (y). All possible time positions for the excitation pulses within a given frame interval are provided. The possible time positions are divided into a number (n.sub.f) of phase positions and each phase position is divided into a number of phases (f). All phases are vacant for the first excitation pulse. When this pulse has been positioned, the phase determined for this pulse is denied to the following excitation pulses until all pulses in a frame have been positioned.

Abstract: A pattern matching system for a speech recognition apparatus includes a dictionary for storing standard speech patterns, where each of the standard speech patterns are stored in the dictionary with two or more different pattern lengths, a converting part for converting an input speech pattern which has an arbitrary pattern length into at least one converted speech pattern which has one of the pattern lengths with which the standard speech patterns are stored in the dictionary depending on the arbitrary pattern length, and a collating part for collating the converted speech pattern with the standard speech patterns stored in the dictionary to obtain a standard speech pattern which has a largest degree of similarity to the converted speech pattern.

Abstract: A voice processing apparatus capable of varying the speed of speech, in which a voice of a predetermined duration is represented by feature parameters and propriety information indicating whether a change in the speech speed is permitted or not. During voice synthesis, the speech speed is varied by skipping or repeating only the feature parameters for which the variation in speech speed is permitted by the associated propriety information.

Abstract: A harmonic coding arrangement where the magnitude spectrum of the input speech is modeled at the analyzer by a relatively small set of parameters and, significantly, as a continuous rather than only a line magnitude spectrum. The synthesizer, rather than the analyzer, determines the magnitude, frequency, and phase of a large number of sinusoids which are summed to generate synthetic speech. Rather than receiving information explicitly defining the sinusoids from the analyzer, the synthesizer receives the small set of parameters and uses those parameters to determine a spectrum, which, in turn, is used by the synthesizer to determine the sinusoids for synthesis.

Abstract: A method of transmitting an audio signal wherein the audio signal is partitioned into successive (in time) blocks by means of time windows, the signal parts contained in the blocks are converted to short-time frequency spectrums by transformation, subsequently the short-time frequency spectrums are coded on the basis of psycho-acoustical masking laws and are transmitted. The received coded signals are decoded, the short-time frequency spectrum signals are brought back into the time domain through re-transformation, and finally the blocks present in the time domain are assembled. Moreover, the blocks are partitioned into sub-blocks and, in case of signal level changes from one block to the other which exceed a predetermined value, the signal parts in the sub-blocks are subjected to a compression prior to transformation and to a complementary expansion after the re-transformation.

Abstract: A bar code reading speech synthesis device is disclosed which includes means for reading a bar code and accessing internal or external memory to supply speech data to a speech synthesizer device based upon the bar code numerical value. Readily and easily changeable memory modules including synthesized speech data developed to accompany corresponding printed lesson plans are also disclosed. The lesson plans, coupled with the speech synthesizer data, provide audible voicing of words and/or audible spelling of the words from the lesson plan. Various other features included in the speech synthesizer device include automatic power down circuitry, self-diagnostics, volume control, and extended diagnostics for testing the functionality of the device. A base.sub.32 addressing scheme is employed to verify memory data as corresponding with bar codes wanded into the device.

Abstract: A first voice recognition apparatus includes a device for analyzing frequencies of the input voice and a device coupled to the analyzing unit for determining vowel zones and consonant zones of the analyzed input voice. The apparatus further includes a device for determining positions of articulation of an input voice determined from the vowel zones by calculating from frequency components of the input voice in accordance with a predetermined algorithm based on frequency components of monophthongs having known phonation contents and positions of articulation. A second voice recognition apparatus includes a device for analyzing frequencies of the input voice so as to derive acoustic parameters from the input voice. A pattern converting unit is coupled to the analyzing unit and uses a neural network for converting the acoustic parameters to articulartory vectors.

Abstract: A temporal sequence of signal vectors X (e.g., generated from a speech signal) is supplied in parallel to elements in an array each of which (after training of the array) provide an output .eta. indicating the vector's similarity to a reference vector. Persistence is built into the elements so that signals forming a "trajectory" within the array, corresponding to an input sequence are simultaneously available for recognition (preferably by a further such array). Arrays may be cascaded to allow for longer sequences.

Abstract: A speech recognition apparatus includes a dictionary for storing information related to registered speeches for use in making a speech recognition, and a registration part for storing the infromation into the dictionary. The registration part includes a filter bank made up of first through nth filters and supplied with a speech which is to be registered in the dictionary, a first circuit part for generating recognition template information based on an output of the filter bank and for storing the recognition template information in the dictionary, and a second circuit part for generating pitch frequency information based on an output of the filter bank and for storing the pitch frequency information in the dictionary. The pitch frequency information is related to a frequency f which satisfies Min.vertline.A(f).vertline., where ##EQU1## X.sub.j (f) denotes a theoretical filter gain of a jth filter of the filter bank at the frequency f, G.sub.

Abstract: In the reporting system disclosed herein, a speech recognizer is used to select sections of text from a report form stored in a computer and to insert recognized terms in the text thereby to generate a report text under voice control. A command interpreter, also responsive to spoken words, initiates creation of the report text and its subsequence storing, printing and transmission. The command processor is responsive to respective spoken commands to select a destination telephone number and to cause the report text to be sent to apparatus for converting report text to image data and for modulating an audio band signal with the image data for facsimile transmission over telephone lines.

Abstract: In an adaptive predictive coding (APC) encoder for compression of quantized digital audio, the signal-to-quantization noise ratio is improved and more complicated input signal spectra can be processed through use of an input signal frequency spectrum detector employing linear predictive coding (LPC) analysis for setting the coefficients of the requantization error noise filter. The detector operates in synchronism with the predictive and noise filters so as to minimize the number of processing calculations while maximizing processing speed.

Abstract: Small light weight voice recorder is worn by a user as a short term memory aid. It has a single, large, readily accessible control button on its front face. The device has two alternating operating modes, each of which is initiated by actuating the same button in the same way. In a first mode the device records a brief spoken message. In the second mode the message is played back. The device has all solid state components with no moving parts for durability, freedom from maintenance and low battery drain. It is especially useful for those with failing memory who may find the complexities and dexterity requirements of conventional recording devices beyond their capabilities. An alternative embodiment of the invention includes covered, rear mounted controls that are easily accessed and operated for playing back the recorded message at a preset time. This feature is designed to not interfere with the primary function of record and playback by alternate operation of a single button.

Abstract: An isolated speech word recognizer for recognizing an input pattern as one of a plurality of known patterns, comprising a similarity information storing unit for storing similarity information representing the degree of significance of a feature in each of the known patterns for recognizing thereof, and a most similar pattern determining unit for determining one of the plurality of known patterns as the most similar pattern to the input pattern by the use of the similarity information, whereby the recognition performance is improved.

Abstract: Coarticulated speech segment data are extracted from spoken carrier syllables and digitally compressed for storage using adaptive differential pulse code modulation (ADPCM). Beginning seed quantization and PCM values are generated for each coarticulated speech segment and stored together with the ADPCM encoded data in a coarticulated speech segment library. ADPCM encoded data are recovered from the coarticulated speech segment library and blown back using the initial quantization and PCM seed values to reconstruct and concatenate in real time the sequence of coarticulated speech segments required by a text to speech program to generate a desired high quality spoken message. In the preferred embodiment of the invention, the coarticulated speech segments are diphones.