Abstract: Disclosed is a method for use by a speech recognizer. The method includes determining a regression class tree structure for the speech recognizer, wherein the tree structure includes, representing word subunits or regression classes, as tree leaves, combining the word subunits to form tree nodes using a distance measure for the word subunits in the acoustic space, and combining regression classes to a regression class that lies closer to a tree root of the tree structure using a correlation measure, and wherein at least two of regression classes having the largest correlation parameter are combined to a new regression class that is used in the formation of the regression tree structure, instead of the two combined regression classes, to determine a regression class representing the tree root.

Abstract: A method for on-line handwriting recognition is based on a hidden Markov model and implies the following steps: sensing real-time at least an instantaneous write position of the handwriting, deriving from the handwriting a time-conforming string of segments each associated to a handwriting feature vector, matching the time-conforming string to various example strings from a data base pertaining to the handwriting, and selecting from the example strings a best-matching recognition string through hidden-Markov processing, or rejecting the handwriting as unrecognized. In particular, the feature vectors are based on local observations derived from a single segment, as well as on compacted observations derived from time-sequential segments.

Abstract: A method and apparatus for processing a sequence of words in a speech signal for speech recognition. The method includes the steps of sampling, at recurrent instants, said speech signal for generating a series of test signals. Signal-by-signal matching and scoring is generated between the test signals and a series of reference signals, where each of the series of reference signals forms one of a plurality of vocabulary words arranged as a vocabulary tree. The vocabulary tree includes a root and a plurality of tree branches wherein any tree branch has a predetermined number of reference signals and is assigned to a speech element and any vocabulary word is assigned to a particular branch junction or branch end. Acoustic recombination determines both continuations of branches and the most probable partial hypotheses within a word because of the use of a vocabulary built up as a tree with branches having reference signals.

Abstract: In the recognition of coherently spoken words, a plurality of hypotheses is usually built up which end in various words during the recognition process and are then to be continued with further words. To keep the number of words yet to be continued as small as possible, especially in the case of a large vocabulary, it is known to carry out a look-ahead in a limited time space. It is suggested according to the invention to use the same phonemes for the look-ahead as for the actual recognition and to add together the differential sums obtained in the look-ahead for the evaluation of the partial hypothesis which has just ended and which is to be continued, and to compare this sum with a threshold value which depends on the extrapolated minimum total evaluation at the end of the time space of the look-ahead. The searching space for hypotheses to be continued can be limited by this in a particularly favorable manner.

Abstract: A speech-actuated control system for use with a consumer product, and in particular, in an automotive vehicle, recognizes a speech command and has a sequencer for after the recognition, effecting a change of state in the control. In particular, a plurality of different commands are recognized. The system has a mechanical actuator for, after the recognition, allowing a specifier command for specifying the most recently recognized speech command. The specifying effects a differential setting change in either of two opposite directions.

Abstract: Reference values are required for deriving word sequences from an unknown speech signal. These reference values are determined from a known test speech signal and represent multi-dimensional mean value vectors which are derived from the characteristic vectors formed during the processing of the test speech signal. As the processing of the test speech signal progresses mean value vectors are continuously split into two new mean value vectors, again. In order to minimize the total number of mean value vectors required for the description of the individual acoustic states of a speech signal, it is proposed to link individual mean value vectors, irrespective of whether or not they are associated with different acoustic states, provided that they neighbor one another closely enough. Notably if the steps of linking and splitting are performed a number of times, preferably in an alternating fashion, there is obtained a set of mean value vectors which approximate the acoustic states comparatively accurately.

Abstract: For speech recognition a new word is represented as based on a stored inventory of models of sub-word units. First a plurality of utterances are presented that all should conform to the word. For building a word model from the utterances, these are represented by a sequence of feature vectors. First, the utterances are used to train a whole-word model that is independent of the models of the sub-word units. The length of the whole-word model equals the average length of the utterances. Next, a sequence of Markov states and associated probability densities of acoustic events of the whole-word model is interpreted as a reference template represented by a string of averaged feature vectors. Finally, the string is recognized by matching to models in the inventory and storing a recognition result as a model of the utterances.