A computer system is described in which isolated words, spoken by a designated talker, are recognized through calculation of a minimum prediction residual. A reference pattern for each word to be recognized is stored as a time pattern of linear prediction coefficients (LPC). The total log prediction residual of an input signal is minimized by optimally registering the reference LPC onto the input autocorrelation coefficients using the dynamic programming algorithm (DP). The input signal is recognized as the reference word which produces the minimum prediction residual. A sequential decision procedure is used to reduce the amount of computation in DP. A frequency normalization with respect to the long-time spectral distribution is used to reduce effects of variations in the frequency response of telephone connections. The system has been implemented on a DDP-516 computer for the 200-word recognition experiment. The recognition rate for a designated male talker is 97.3 percent for telephone input, and the recognition time is about 22 times real time.

Minimum prediction residual principle applied to speech recognition

The first edition of this book has enjoyed a gratifying existence. 1s sued in 1965, it found its intended place as a research reference and as a graduate-Ievel text. Research laboratories and universities reported broad use. Published reviews-some twenty-five in number-were universally kind. Subsequently the book was translated and published in Russian (Svyaz; Moscow, 1968) and Spanish (Gredos, S.A.; Madrid, 1972). Copies of the first edition have been exhausted for several years, but demand for the material continues. At the behest of the publisher, and with the encouragement of numerous colleagues, a second edition was begun in 1970. The aim was to retain the original format, but to expand the content, especially in the areas of digital communications and com puter techniques for speech signal processing. As before, the intended audience is the graduate-Ievel engineer and physicist, but the psycho physicist, phonetician, speech scientist and linguist should find material of interest."

Speech Analysis, Synthesis and Perception

A system D interfaces with a multiplicity of individual terminals T1-Tn of a telephone network facility C, at the terminals callers are prompted by voice-generated instructions to provide digital data that is identified for positive association with a caller and is stored for processing. The caller's identification data is confirmed using various techniques and callers may be ranked and accounted for on the basis of entitlement, sequence or demographics. Callers are assigned random designations that are stored along with statistical and identification data. A break-off control circuit may terminate the computer interface aborting to a terminal for direct communication with an operator. Real-time operation processing is an alternative to stored data. The accumulation of stored data (statistical, calling order sequence, etc.) is variously processed and correlated as with developed or established data to isolate a select group or subset of callers who can be readily identified and reliably confirmed. Different program formats variously control the processing of statistical data as for auction sales, contests, lotteries, polls, commercials and so on.

Telephonic-interface statistical analysis system

Minimum Prediction Residual Principle Applied to Speech Recognition

An improved noise suppression system (400) is disclosed which performs speech quality enhancement upon speech-plus-noise signal available at the input (205) to generate a clean speech signal at the output (265) by spectral gain modification. The noise suppression system of the present invention includes a background noise estimator (420) which generates and stores an estimate of the background noise power spectral density based upon pre-processed speech (215), as determined by the detected minima of the post-processed speech energy level. This post-processed speech (255) may be obtained directly from the output of the noise suppression system, or may be simulated by multiplying the pre-processed speech energy (225) by the channel gain values of the modification signal (245). This technique of implementing post-processed signal to generate the background noise estimate (325) provides a more accurate measurement of the background noise energy since it is based upon much cleaner speech signal. As a result, the present invention performs acoustic noise suppression in high ambient noise backgrounds with significantly less voice quality degradation.

Noise suppression system

A plurality of pitch period candidates are selected from a peak of correlation of a speech waveform in a current frame from which a pitch period is to be extracted, and a speech pitch is selected from the candidates by referring to a guide index which is precalculated based on pitch periods extracted in past frames. The guide index is an average of the pitch periods in the past frames.

Method and apparatus for extracting speech pitch

The generation of nasal consonants can be simulated approximately by a cascade connection of simple electrical tuned circuits excited by a quasi‐periodic electrical buzz source. The lowest resonance for nasal consonants is in the vicinity of 200 to 300 cps, and the damping of this resonance is greater than that for the first resonance of vowels. A terminal‐analog synthesizer and control device have been used to generate a number of synthetic syllables, each consisting of a nasal consonant followed by a vowel, with smooth formant transitions between consonant and vowel portions. Systematic listening tests have shown that the identification of the nasal consonant is determined largely by the frequency position of the second resonance of the nasal portion of the syllable, and thus by the direction and extent of the second formant transition of the vowel, as noted in other studies. The identification of the nasal consonant is also dependent to some extent on the duration of the consonant and transition portions of the syllable. Isolated nasal consonants generated by this procedure can also be identified with reasonable consistency.

Synthesis and Perception of Nasal Consonants

Various parameter sets-including a spectrum envelope, cepstrum, autocorrelation function, linear predictive coefficients, and partial autocorrelation coefficients (PAC's)- are evaluated experimentally to determine which constitutes the best parameter in spoken digit recognition. The principle of recognition is simple pattern matching in the parameter space with nonlinear adjustment of the time axis. The spectrum envelope and cepstrum attain the best recognition score of 100 percent for ten spoken digits of a single-male speaker. PAC's seem to be preferable because of their ease of extraction and theoretical orthogonalities; however, these PAC's tend to suffer from computation errors when computed by fixed-point arithmetic with a short accumulator length. We find two effective means to improve the errors; one is variable use of the PAC dimensions controlled by computation accuracy, and the other is smoothing along the time axis. With these improvements the PAC's offer almost 100 percent recognition.

https://ieeexplore.ieee.org/iel6/8337/26098/01162480.pdf

Evaluation of various parameter sets in spoken digits recognition

Signal-to-noise is enhanced by including the upper sideband of a sampled-speech signal with the original baseband signal for further signal processing. The invention features shifting both bands to form a continuum from 0 Hz to the sampling frequency. Application in a speech recognition is shown.

Preprocessing method and device for speech recognition device

A method and apparatus for speech signal detection and classification in which a partial auto-correlation and residual power analyzation circuit extracts a normalized first-order partial auto-correlation coefficient and K 1 a normalized zero-order residual power E N from an input signal, and a sound source analyzation circuit extracts a normalized residual correlation φ from the input signal, and in which on the basis of these extracted parameters, speech signals are detected, and, when so detected, the detected speech signals are classified into a voiced sound V, an unvoiced sound U and silence S. The classification of the respective voiced sound, unvoiced sound and silence is determined on the basis of preset threshold values that are mutually considered and which correspond to values of these extracted K 1 , E N and φ parameters for establishing boundary values for classifying the input signals into a voiced sound, an unvoiced sound or silence.

Kazuo Nakata

Papers

Method and apparatus for extracting speech pitch

Synthesis and Perception of Nasal Consonants

Evaluation of various parameter sets in spoken digits recognition

Preprocessing method and device for speech recognition device

Method and apparatus for speech signal detection and classification of the detected signal into a voiced sound, an unvoiced sound and silence