Oki Electric Industry Co., Ltd.

Waseda University

Makoto Morito

Yukio Tabei

Takashi Yazu

Kozo Yamada

Kenzo Akagiri

Akihiko Fujisawa

Keiko Takahashi

Kei Yamada

Tetsunori Kobayashi

Masao Takeuchi

Tetsuji Ogawa

In a source sound separator, first and second target sound predominant spectra are generated respectively by first and second processing operations for linear combination for emphasizing the target sound, using received sound signals of two microphones arrayed at a distance from each other. A target sound suppressed spectrum is generated by processing for linear combination for suppression of the target sound, using the two received sound signals. Further, a phase signal containing a larger amount of signal components of the target sound and exhibiting directivity in the direction of the target sound is generated by processing of linear combination, using the two received sound signals. The target sound and the interfering sound are separated from each other using the first and second target sound predominant spectra, the target sound suppressed spectrum, and the phase signal.

Source sound separator with spectrum analysis through linear combination and method therefor

A noise suppressor detects a peak position in the frequency spectrum of an input speech signal, and masks frequency components in the spectrum as a function of the peak position. The masking process attenuates or removes frequency components near the peak position if their magnitudes are significantly lower than the magnitude of the spectrum at the peak position. This noise suppressor effectively removes irregular noise from the spectrum while leaving enough of the spectrum to reproduce the speech signal clearly.

Noise suppressor for removing irregular noise

A speech recognition apparatus is adapted to previously prepare a noise pattern in response to environmental noise prior to inputting a speech signal, evaluate a speech feature vector B.sub.i yielded by subtracting the noise pattern from a feature vector A.sub.i of the input speech upon inputting the speech signal thereafter, spectrum-normalize the speech feature vector B.sub.i, evaluate a local peaks vector by making use of binary-coding processing wherein only a component of a channel being the maximum of the spectrum-normalized vector in a direction of frequency is assumed to be '1', evaluate pattern similarity between an input pattern comprising a local peaks vector from the start point to the end point of the input speech and previously prepared reference patterns of a plurality of categories of the same format as the input pattern, and judge a category of a reference pattern being the maximum among the pattern similarities as a recognized result. As a result, any influence of environmental noise on the speech of concern is eliminated for assuring highly accurate recognition. Furthermore, the speech recognition apparatus is adapted to improve the efficiency and accuracy of speech recognition by evaluating a new speech feature vector of a prescribed frame number by effecting linear expansion therefor on the basis of internal division for adjoining speech feature vectors B.sub.i and B.sub.i +1, and smoothing a local peaks vector.

Speech recognition apparatus for recognizing the category of an input

In a speech recognition system for effecting substantially linear matching, a start point of an input speech pattern is detected to start counting the frames of the input speech pattern for updating a speech frame number. Each time the speech frame number is updated, a plurality of frame numbers of reference templates are generated in substantially a linear relationship to the frame number, thus providing a plurality of matching paths between each of the reference templates and the input speech pattern. The distance between the input pattern data and the reference template data between the corresponding frames specified by each matching path is calculated each time the speech frame number is updated. An accumulated value of the distances along a matching path from the start point of the input speech until a desired speech frame is regarded as a dissimilarity, and the dissimilarity corresponding to each matching path in each reference template is calculated each time the speech frame is updated. The number of the reference template which gives a minimum dissimilarity is detected and stored. At the time of confirming an end of the input speech, the number of the reference template with the minimum dissimilarity is determined to be the recognized result of input speech.

Apparatus and method for recognizing speech

In a sub-band speech analyzing and synthesizing apparatus, a low-pass filter comprises a nonrecursive filter. The center frequency and bandwidth of each channel are selected so that the decimated sampling period of that channel is an integer multiple of the period of the modulating signal or the demodulating signal of the channel. Modulation or demodulation is performed simultaneously with the low-pass filtering by the nonrecursive filter.

Sub-band speech analyzing and synthesizing device

Speech recognition method and system are adapted to previously prepare a noise pattern in response to environmental noise prior to inputting a speech signal, evaluate a speech feature vector Bi yielded by subtracting the noise pattern from a feature vector Ai of the input speech upon inputting the speech signal thereafter, spectrum-normalize the speech feature vector Bi, evaluate a local peaks vector by making use of binary-coding processing wherein only a component of a channel being the maximum of the spectrum-normalized vector in a direction of frequency is assumed to be '1', evaluate pattern similarity between an input pattern comprising a local peaks vector from the start point to the end point of the input speech and previously prepared reference patterns of a plurality of categories of the same format as the input pattern, and judge a category of a reference pattern being the maximum among the pattern similaritites as a recognized result. As a result, any influence of environmental noise on the speech in concern is eliminated for assuring highly accurate recognition. Furthermore, the speech recognition method and system are adapted to improve efficiency and accuracy of speech recognition by making use of processing of evaluating a new speech feature vector of a prescribed frame number of effecting linear expansion therefor on the basis of internal division of adjoining speech feature vectors Bi and Bi+1, and processing of smoothing a local peaks vector.

Speech recognition for recognizing the catagory of an input speech

A speech synthesizer for reproducing natural speech with small memory. Synthesized speech consists of symmetric speech segments. Only half of a symmetric speech segment is stored in a memory through ADPCM code, and the other half of the segment is generated by using the stored half segment. Repeat control data is also stored in a memory so that only a single segment is enough to be stored when the same segments are repeated. In the repeated process, the amplitude of the segment is changed smoothly so that amplitude discontinuity between the two successive segments does not occur. The synthesized digital speech data is converted to an analog waveform by using a digital to analog converter.

Growing community of inventors

Tokyo, Japan

Makoto Morito