Abstract: A binary encoder for vector quantization is provided which comprises a plurality of identical two-level branch selectors connected in a turnaround cascade pipeline array. The upper levels of the two-level selectors are connected in series and the first selector receives a formatted digital data vector input. The upper level of last selector has its output connected to its own lower level input and the outputs of the lower level selectors are connected in series so that the last lower level selector in the turnaround cascade resides in the first two level selector. The output of the last lower level selector provides a desired compressed data vector output.

Abstract: Various selected characteristics of digital (or analog) signals in an electrical circuit are converted into audible signals and transmitted as a sound pattern(s) through a speaker or headphones to a user. Such selected signal characteristics include pulse width, duty cycle, activity level (state changes), repetitive pattern comparison, and sampled signal level. Some representative types of differentiable noise or sounds include clicks, pitch, harmonics, noise bursts, ticks and tocks, and voice stopped consonants. By matching selected noise types with selected characteristics an audible display is generated which represents to the user various characteristics of the sampled signal either individually or collectively.

Abstract: The method disclosed consists in analyzing the vocal signals in consecutive windows, in quantifying, on a determined number m of levels, the vocal signal in each of the windows and in measuring, in each of the windows, the vocal signal in each of the windows and in measuring, in each of the windows, the r-m-s value of the samples of the vocal signal. It consists constructing, in a vector space with n dimensions having, as its first base, the unit vectors (E.sub.0 to E.sub.n-1) of the energies measured on n consecutive windows, a resultant energy vector E corresponding to the sum of n energy vectors measured respectively in n windows of analysis of the vocal signal, then in achieving, in this space, a change of base having, as its first main axis, an oriented axis of unit vector having, as its components, the unit vectors of the first base, to project, in the new base obtained, the resultant energy vector. Codings on q bits such that 2.sup.

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: A method and apparatus for removing the periodicity from a speech signal in a transform coder prior to the quantization of the speech signal, which speech signal is a sampled time domain speech signal composed of information samples, the transform coder sequenctially segregating the speech signal into blocks of information samples, is shown to include apparatus and method for determining the pitch in each of the sample blocks, determininig a long term predetermined parameter (LTP) for each of the blocks based on the pitch determined for each block, calculating a periodicity value for each sample in the block wherein the calculation of the periodicity value is based upon the pitch and the long term predictor parameter, generating a revised block of difference samples by subtracting the periodically value from the corresponding sample, and performing adaptive transform coding on each of the difference blocks.

Abstract: Noise in a speech-plus-noise input signal is suppressed by splitting the input signal into spectral channels and decreasing the gain in the each channel which has a low signal-to-noise ratio (SNR). A voice operated switch (VOX) acts to detect noise-only input to gate a background noise (input signal) estimator and also to gate a residual noise (output signal) estimator. The gain in each of the channels is controlled by the current value (a posteriori) input signal SNR estimate, modified by the prior value (a priori) input signal SNR estimate, and smoothed as a function of the residual (output noise signal) estimate.

Abstract: In an arrangement for coding multi-element signals such as used in speech or image processing, a plurality of N element reference signals representable in a prescribed vector space are stored. An N element input signal representable in the prescribed vector space is received and one of the reference signals is selected to represent the input signal. A set of signals each representative of the projection of one of the reference signals on a predetermined orientation in the prescribed vector space and a signal representative of the projection of the input signal on the predetermined orientation are formed. The reference signals are arranged according to their projections on the predetermined orientation.

Abstract: An interactive computer assisted language learning system which allows a student to select a model phrase from text displayed on an electronic display; record (in digitized form) his own pronunciation of that phrase; and instantly listen to the digitized vocal version of the selected phrase and his own recorded pronunciation for comparison purposes. An audio CLIP mode permits the student to select any (random) portion of displayed text (e.g., a phrase, a small part of a phrase, a single word, a syllable, or a phoneme) using cursor control or the like and to control the system to play the voice corresponding to that selected portion. A SoundSort text reconstruction exercise based on aural clues automatically randomizes the order of plural phrases, provides digitized utterances of the phrases in the randomized order, and requires the student to reconstruct the original order using a visual display interface.

Abstract: A speech recognition method and apparatus take into account a system transfer function between the speaker and the recognition apparatus. The method and apparatus update a signal representing the transfer function on a periodic basis during actual speech recognition. The transfer function representing signal is updated about every fifty words as determined by the speech recognition apparatus. The method and apparatus generate an initial transfer function representing signal and generate from the speech input, successive input frames which are employed for modifying the value of the current transfer function signal so as to eliminate error and distortion. The error and distortion occur, for example, as a speaker changes the direction of his profile relative to a microphone, as the speaker's voice changes or as other effects occur that alter the spectra of the input speech frames. The method is automatic and does not require the knowledge of the input words or text.

Abstract: A system for storing and forwarding voice signals. The system provides for central, digital storage of voice signals for later access by addressee system users. When addresses access a previously stored voice signal they have the capability to incrementally construct a reply by "toggeling" between a playback state, to listen to the previously stored signal and a record state to record at least a partial response. In another embodiment originators of voice signals have the capability to edit signals after storage.

Abstract: The methods and apparatus disclose a signal processing system acquiring the half-period and magnitude of the highest frequency component at any one time of an analog signal. Two comparators compare positive and negative going slopes of the signal to respective out of phase versions of themselves. Maxima and minima are detected by the respective comparators to set and reset two timers. The timers time the lengths of the positive and negative going slopes between the maxima and minima. An analog to digital converter converts the magnitude of the signal at the maxima and minima. A microprocessor stores the times and magnitudes in a memory and is in a second embodiment adapted to determine the individual frequency components of the signal from the stored values. The acquired values may be transmitted in digital form or may be reconstructed for analog transmission.

Abstract: This telephone solicitation system provides automatic dialing and call-progress detection. Audio signals on the line are analyzed to recognize network operation tones, noise bursts, and speech. A "sample" is the total number of crossings during a 25 millisecond period. A "window" is a sequence of eight samples analyzed for spectral and time patterns.

Abstract: Statistically analyzing a discriminant variable generated by a discriminant voiced detector is done to determine the presence of the fundamental frequency in a changing speech environment. The detector is responsive to the discriminant variable to first calculate the average of all of the values of the discriminant variable over the present and past speech frames and then to determine the overall probability that any frame will be unvoiced. In addition, the detector calculates two values, one value represents the statistical average of discriminant values that an unvoiced frame's discriminant variable would have and the other value represents the statistical average of the discriminant values for voice frames. These latter calculations are performed utilizing not only the average discriminant value but also a weight value and a threshold value which are adaptively determined from frame to frame. The unvoiced/voiced decision is made by utilizing the weight and threshold values.

Abstract: A pulse train of primary pulses is estimated from an inverse LPC analysis of a frame of voiced speech. From this estimated pulse train a pole-zero filter is estimated. The estimated pulse train is used to excite the estimated pole-zero filter to produce a synthesized speech signal. The synthesized speech signal is compared to the original frame of speech to determine the error in the original speech signal. Both the pulse amplitude and filter are adjusted to compensate for the error and another synthesized speech signal is produced. The process may be repeated until the synthesized speech signal and original speech signal converge.

Abstract: This invention utilizes a time-varying recursive filter where the multipliers following successive delay elements of the filter have sets of normalized covariance matrix coefficients which are stored and which have been obtained from the normalized autocorrelation coefficients for each of a plurality of time frames of the sampled transient signal which is to be synthesized from the stored coefficients. The normalization constant during a frame is also applied as a scale factor to the recursive filter. The input of the filter is a pseudo-random noise generator signal applied to the recursive filter at the sample rate. A plurality of successive time frames of operation of the recursive filter and with a set of coefficients for each time frame provides the entire synthesized transient signal. An analog and digital implementation of the synthesizer are described.

Abstract: A speech recognition method and apparatus in which a plurality of time-sequential acoustic parameters are derived from a segmented input speech signal and are used to form a first trajectory that is time normalized. The time-normalized trajectory is sampled at predetermined lengths therealong and the sampling results used to form a new time-normalized trajectory with equally spaced points ("dots" on the graph), which is compared with a plurality of previously registered trajectories until a match is found between the new time-normalized trajectory and one of the registered trajectories, at which time an indication of the results of the matching is produced. A silence acoustic parameter can also be added to the time-sequential acoustic parameters so that the time-normalized trajectory can start and end from a point of silence.

Abstract: A voice coding apparatus includes a pitch detecting circuit which detects a pitch period of a voice signal; a pitch waveform generating circuit which samples the voice signal for a plurality of pitches based on the pitch period detected by the pitch detecting circuit and which generates a waveform of one pitch from the waveform of the plurality of pitches; a band restriction circuit which restricts the frequency band of the one pitch waveform generated in the pitch waveform generating circuit; and a coding circuit for coding the voice waveform which is band restricted in the band restriction circuit. The sampling number of the waveform for a plurality of pitches and the restricted bandwidth can be changed in accordance with the amount of the pitch period extracted in the pitch detecting circuit. Further, the pitch detecting circuit is able to correctly detect the pitch period even when the pitch period is not a multiple of the sampling period.

Abstract: A method and apparatus are disclosed for recognizing spoken commands uttered by a user and for generating responsive control signals once the command is recognized. In accordance with this disclosure the audio signal is converted into a series of count bytes representing the time between the audio signal zero crossings, and all the count bytes of the full command are then segmented into equal temporal groups histogram and sorted within each segment into a set of frequency class intervals which are based on a computation of substantially equal byte activity in all the words comprising the command lexicon. In this manner, lower and higher frequency groups are selected for equal significance. The uttered words are then compared against stored words similarly transformed according to segment and frequency interval and if the comparison conditions are satisfied the command is executed; if not, an indication is provided to the user to repeat the command.

Abstract: A speech recognition LSI system has a first sampling circuit which samples, at a first frame interval, which is shorter than 1/2fc (fc denotes the cut-off frequency of a low-pass filter in a speech processing section), an A/D conversion output signal in each frequency band of a speech input signal. Thereafter, the sampled data of a plurality of frames is stored in a digital low-pass filter where is smoothed. Output data which has been smoothed by the digital low-pass filter is then supplied to a second sampling circuit which in turn produces sampled data having a second frame interval, this being longer than 1/2fc. The sampled data sequentially supplied from the second sampling circuit is supplied to a recognition-processing unit to detect the speech segment. The matching degree between data in the detected speech segment and reference pattern data is detected and the recognition result for the speech input signal is derived out on the based of the magnitude of the similarities detected.

Abstract: A speech recognizer includes a plurality of stored constrained hidden Markov model reference templates and a set of stored signals representative of prescribed acoustic features of the said plurality of reference patterns. The Markov model template includes a set of N state signals. The number of states is preselected to be independent of the reference pattern acoustic features and preferably substantially smaller than the number of acoustic feature frames of the reference patterns. An input utterance is analyzed to form a sequence of said prescribed feature signals representative of the utterance. The utterance representative prescribed feature signal sequence is combined with the N state constrained hidden Markov model template signals to form a signal representative of the probability of the utterance being each reference pattern. The input speech pattern is identified as one of the reference patterns responsive to the probability representative signals.