Abstract: The text is generated from voice input that divides the processing of each spoken word into a dictation event and a text event. Each dictation event handles the processing of data relating to the input into the system, and each text event deals with the generation of text from the inputted voice signals. In order to easily distinguish the dictation events from each other and text events from each other the system and method creates a data structure for storing certain information relating to each individual event. Such data structures enable the system and method to process both simple spoken words as well as spoken commands and to provide the necessary text generation in response to the spoken words or to execute an appropriate function in response to a command.

Abstract: In the speech recognition system disclosed herein, an input utterance is submitted to both a large vocabulary isolated word speech recognition module and a small vocabulary continuous speech recognition module. The small vocabulary contains command words which can be combined in sequences to define commands to an application program. The two recognition modules generate respective scores for identified large vocabulary models and for sequences of small vocabulary models. The score provided by the continuous speech recognizer is normalized on the basis of the length of the speech input utterance and an arbitration algorithm selects among the candidates identified by the recognition modules. Without requiring the user to switch modes, text is output if a score from the isolated word recognizer is selected and a command is output if a score from the continuous speech recognizer is selected.

Abstract: A speech recognition system is disclosed which comprises a core speech recognition program and a plurality of utility program modules for adjusting various recognition parameters such as gain, sensitivity and acceptance threshold and for improving the training of word models. The system further provides a decision tree and utility controlling program module which can be invoked by a user confronting problems during the running of the core program. The system utilizes user input to traverse the decision tree and to initiate appropriate ones of the utility program modules thereby to alter the on-going behavior of the core recognition program.

Abstract: In the large vocabulary speech recognition system disclosed herein, a preliminary screening of vocabulary models is provided by applying high speed distance measuring functions. The distance measuring functions utilize subsampled or otherwise reduced representations of the unknown speech segment and the vocabulary models. The initial screening functions achieve very high speed by precalculating, for each utterance, a comparison table of distance values which can be used for all vocabulary models. The building of each comparison table is facilitated by a method which utilizes default values as initial entries and only adjusts entries which are meaningfully different from the default value.

Abstract: In the speech recognition system disclosed herein, an input utterance is submitted to both a large vocabulary isolated word speech recognition module and a small vocabulary continuous speech recognition module. The two recognition modules generate respective scores for identified large vocabulary models and for sequences of small vocabulary models. The score provided by the continuous speech recognizer is normalized on the basis of the length of the speech input utterance and an arbitration algorithm selects among the candidates identified by the recognition modules. Preferably, the competing scores from the two recognizers are scaled by a factor or factors empirically trained to minimize incursion by each of the vocabularies on correct results from the other vocabulary.

Abstract: The electronic dictionary disclosed herein is organized for expeditious search based on partial spelling by assigning words to blocks having a predetermined maximum size, the blocks being represented by respective partial spelling sequencer. The words are assigned to blocks by progressing through successive possible sequences in order and, for each sequence, determining the number of words in the set of words corresponding to that sequence. If the number of words is less than the maximum, all of the words in the set are assigned to a corresponding terminal block. Otherwise words up to a preselected number are assigned to a non-terminal block and the partial spelling sequence is extended. As a result of the organization only one call to the dictionary needs to be made for each extension of the sequence.

Abstract: The speech recognition system disclosed herein obtains improved recognition accuracy by employing recognition models which are discriminatively trained from a data base comprising training data from different sources, e.g., both male and female voices. A linear discriminant analysis is performed on the training data using expanded matrices in which sources are identified or labelled. The linear discriminant analysis yields respective transforms for the different sources which however map the different sources onto a common vector space in which the vocabulary models are defined.

Abstract: An input utterance is converted to a sequence of standard or prototype data frames which are compared with word models which are represented by respective sequences of standard or prototype probability states, there being a pre-calculable distance metric representing the degree of match between each prototype data frame and each prototype model state. Only distance measurements better than a calculated threshold are considered meaningful and those meaningful metrics are stored in a packed list. Also stored is an address array of offsets for locating particular meaningful metrics in the list, the address array being accessed by the corresponding frame and state indices. Also stored is an array for distinguishing meaningful and non-meaningful metrics. Accordingly, an input utterance can be evaluated by locating meaningful metrics in the packed list using the address array and by utilizing a default value for any non-meaningful metric.

Abstract: The method disclosed herein facilitates the generation of a recognition model for a non-standard word uttered by a user in the context of a large vocabulary speech recognition system in which standard vocabulary models are represented by sequences of probability distributions for various acoustic symbols. Along with the probability distributions, a corresponding plurality of converse probability functions are precalculated which represent the likelihood that a particular probability distribution would correspond to a given input acoustic symbol. For a non-standard word uttered, a corresponding sequence of acoustic symbols is generated and, for each such symbol in the sequence, the most likely probability distribution is selected using the converse probability functions.

Abstract: Preliminary screening of vocabulary models is provided by successively applying two different high speed distance measuring functions which provide progressively increasing measurement accuracy. Both distance measuring functions utilize subsampled representations of the unknown speech segment and the vocabulary models. The initial screening function achieves very high speed by eliminating certain usual time warping constraints and by precalculating a table of distance values which can be used for all vocabulary models. The second screening function yields improved accuracy in spite of possible endpointing errors by comparing extra frames, preceding and following the presumed unknown word, with noise models appended to each vocabulary model.

Abstract: In the speech recognizer disclosed herein, alignment of an unknown speech sediment, represented by a finely gradiated sequence of frames, with a model sediment represented by a sequence of states is performed by first preparing respective coarse sequences representing the unknown and model segments thereby to define a coarse matrix representing possible alignments. The fine sequences correspondingly define a fine matrix. A best alignment of the coarse sequences is determined thereby to define a coarse path through the coarse matrix. The coarse path is overlaid on the fine matrix and a corridor is defined which includes fine matrix locations which lie within a preselected metric of the coarse path. Only transitions within the corridor are calculated in determining the fine alignment of the unknown speech segment with the model segment, thereby significantly reducing the number of computations required.

Abstract: In a continuous speech recognizer which includes at least, one acoustic expert and one linguistic expert which generate respective scores, a method is disclosed for adjusting the relative weighting to be applied to those scores employing training data utilizing the words to be recognized in multiple word phrases. Multiple word test phrases are applied to the acoustic expert to determine, for each phrase, plural multi-word hypotheses each having corresponding cumulative scores. The linguistic expert generates corresponding cumulative linguistic scores. An objective function is calculated for each test phrase having a value which is variable as a function of the difference between the combined score of any correct hypothesis and that of the most easily confused incorrect hypothesis. The objective function values are cumulated and a gradient descent procedure is used to adjust the relative weighting of the acoustic and linguistic scores in obtaining a combined score.

Abstract: Disclosed is a system and method for generating text from a voice input that divides the processing of each speech event into a dictation event and a text event. Each dictation event handles the processing of data relating to the input into the system, and each text event deals with the generation of text from the inputted voice signals. In order to easily distinguish the dictation events from each other and text events from each other the system and method creates a data structure for storing certain information relating to each individual event. Such data structures enable the system and method to process both simple spoken words as well as spoken commands and to provide the necessary text generation in response to the spoken words or to execute an appropriate function in response to a command. Speech recognition includes the ability to distinguish between dictation text and commands.

Abstract: In the reporting system disclosed herein, a speech recognizer is used to select sections of text from a report form stored in a computer and to insert recognized terms in the text thereby to generate a report text under voice control. A command interpreter, also responsive to spoken words, initiates creation of the report text and its subsequence storing, printing and transmission. The command processor is responsive to respective spoken commands to select a destination telephone number and to cause the report text to be sent to apparatus for converting report text to image data and for modulating an audio band signal with the image data for facsimile transmission over telephone lines.

Abstract: The speech recognition system disclosed herein operates to select, from a collection of tokens which represent vocabulary words, those tokens which most closely match an unknown spoken word. The collection of tokens is divided into partitions, each of which is characterized or identified by a representative one of the tokens. Both the tokens and the unknown speech word are represented by a sequence of standard data frames which may, for example, define characteristic spectra. In operation, the system computes the distance from the unknown to each of the representative tokens and then, starting with the partition having the nearest representative token and proceeding through partitions represented by successively more distant tokens, examines the other tokens in that partition while keeping a list of predetermined length identifying the examined tokens which thus far provide the best match. This process is continued until the number of distance calculations performed reaches a preselected level.

Abstract: A speech recognition apparatus having reference pattern adaptation stores a plurality of reference patterns representing speech to be recognized, each stored reference pattern having associated therewith a quality value representing the effectiveness of that pattern for recognizing an incoming speech utterance. The method and apparatus provide user correction actions representing the accuracy of a speech recognition, dynamically, during the recognition of unknown incoming speech utterances and after training of the system. The quality values are updated, during the speech recognition process, for at least a portion of those reference patterns used during the speech recognition process. Reference patterns having low quality values, indicative of either inaccurate representation of the unknown speech or non-use, can be deleted so long as the reference pattern is not needed, for example, where the reference pattern is the last instance of a known word or phrase.

Abstract: A report generation method and apparatus automatically bind and capitalize text strings provided, for example, by speech recognition systems as text to be inserted into a report form. The method and apparatus associate with each text string, text string signals representing instructions for binding and capitalizing the text strings. Spaces, if any, are then prepended to the next string in response to the text string signals and the first alphanumeric letter of the text string is capitalized in response to the text string signals. The text string signals include left and right codes or instructions describing the beginning and ending of the text material and can include either preset or machine-generated action instructions designating the binding and capitalization requirements for the text string. The apparatus and method preferably employ a look-up table for associating the binding and capitalization requirements of the text string with the text string code types.

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: In a speech recognition system disclosed herein, acoustic speech waveforms are initially analyzed to obtain, at successive sample times, digital frames of speech information. This initial analysis may, for example, be performed by multi-channel filtering or linear predictive encoding. Stored in the apparatus is a list of representative standard frames, represented by coded indices, together with a table of difference values which represent the vector distances between each standard frame in the list and all other standard frames. For each token (vocabulary) word which is to be recognized, there is stored a sequence of standard frame indices which represent that token word. As each sample frame is generated, a representative standard frame is selected which best represents the sample frame.