Threshold Technology, Inc.

Marvin B Herscher

Thomas B Martin

Gene Grunza

Robert B Cox

John R Welch

In this speech recognition system the array formed by a timewise sequence of speech signal feature vectors includes digital data at each time slot representing both presence/absence and consistency of occurence.

Speech recognition apparatus and method

An apparatus is disclosed for receiving and processing transmitted electromagnetic waves, for example transmitted radio signals which contain audio information including recognizable speech information. In accordance with the invention, there are provided first and second spaced antennas. Switching means are provided for coupling either the first or the second antenna to a common channel. Signal processing means are provided for processing the signals on the common channel. Means are provided for comparing the signal on the common channel with a predetermined threshold condition and for generating a control signal in response to the comparison. Finally, means are provided for applying the control signal to the switching means to control selection of the first or second antenna. In one preferred embodiment of the invention, the comparator means comprises means for comparing the energy level of the signal at the common channel with a predetermined energy threshold. During operation, the antenna which is active at a particular time will remain active (i.e. coupled to the common channel) so long as the received radio frequency energy, at the characteristic transmission frequency, remains above the predetermined threshold level. If the level of received energy at this characteristic frequency falls below the predetermined threshold level (for example, by virtue of the antenna momentarily being in a multipath 'dead' spot), the comparator means will sense this condition and operate the switch such that the other antenna is coupled to the common channel and becomes active. In another form of the invention the signal processing means includes means for generating an automatic gain control (AGC) signal which depends upon the instantaneous level of the signal on the common channel. A signal derived from the AGC signal is compared with a threshold signal that depends upon a predetermined reference, and a control signal is generated when the signal derived from the AGC signal exceeds the threshold signal.

Receiving apparatus having a plurality of antennas

An apparatus for recognizing the occurrence of a command word within continuous speech, features an improved sequential processing of feature signals derived from the input speech: feature subsets are compared with previously stored subset signals to determine the time interval or boundary of command word candidates. The occurrence decision and indication of the command word is made from a comparison of a feature signal matrix versus a previously stored training matrix.

Apparatus for recognizing words from among continuous speech

An apparatus which receives spoken input training words and a subsequent spoken input command word and generates a correlation function is indicative of the resemblance of the command word to each training word. A feature extraction means processes received input words and generates digital feature output signals on particular ones of a number of feature output lines. Means are provided for storing, as a time dependent matrix, the status of the feature signals which occur during each training word. Means are also provided for storing as a time dependent matrix, the status of the feature signals which occur during the command word. Further means are provided for comparing, member by member, the command word matrix with each training word matrix and for generating a correlation figure which reflects each comparison. In a preferred embodiment of the invention, means are provided for time normalizing the matrices representative of the training words and the command word before the comparisons are made.

Word recognition apparatus and method

The present invention pertains to an apparatus which receives acoustic input, the input including words spoken in isolation, finds the word boundary instants at which a word begins and ends, and performs recognition functions on the words. A feature of the invention is the compensation for breath noise after the true end of a word, using a variable backup of the estimated word end. The apparatus includes means for generating feature signals indicative of feature characteristics in the received input and further includes means for comparing the feature signals which occurred during determined time boundaries with stored features corresponding to words in a vocabulary. The invention is directed to an improved system for detecting word boundaries which includes a means responsive to the input for generating a first feature signal indicative of the substantially continuing presence of speech-like sounds which meet a first selection criterion. Means are provided for storing the feature signals which occur during the presence of this first feature signal. Further means, responsive to the input, are provided for generating a second feature signal indicative of the presence of speech-like sounds which meet a second selection criterion, this second selection criterion being more restrictive than the first selection criterion. Means are also provided for determining the last occurrence of the second feature signal among the stored feature signals. The end boundary of an input spoken word is determined as a function of this last occurrence.

Word boundary detector for speech recognition equipment

An apparatus and method for recognizing the occurrence of a command word from among input speech that may be continuous in form. In accordance with the invention, means are provided for generating feature signals which depend on the features of an input word being spoken. The feature signals are processed to determine the time interval of occurrence of a predetermined sequence of features. Also provided is a means for comparing the feature signals which occur during the determined time interval with a stored set of features that are expected to occur characteristically during the command word to determine the degree of correlation therebetween.

Growing community of inventors

Cherry Hill, NJ, United States of America

Marvin B Herscher