Abstract: Voice utterances are substituted for manipulation of a pointing device of the kind which is manipulated to control motion of a cursor on a computer display and to indicate desired actions associated with the position of the cursor on the display. Voiced utterances are converted to commands, expressed in a predefined command language, to be used by an operating system of a computer, by converting some voiced utterances into commands corresponding to actions to be taken by the operating system, and converting other voiced utterances into commands which carry associated text strings to be used as part of text being processed in an application program running under the operating system. A table is generated for aiding the conversion of voiced utterances to commands for use in controlling an operating system of a computer to achieve desired actions in an application program running under the operating system.

Abstract: Voice utterances are substituted for manipulation of a pointing device, the pointing device being of the kind which is manipulated to control motion of a cursor on a computer display and to indicate desired actions associated with the position of the cursor on the display, the cursor being moved and the desired actions being aided by an operating system in the computer in response to control signals received from the pointing device, the computer also having an alphanumeric keyboard, the operating system being separately responsive to control signals received from the keyboard in accordance with a predetermined format specific to the keyboard; in the system, a voice recognizer recognizes the voiced utterance, and an interpreter converts the voiced utterance into control signals which will directly create a desired action aided by the operating system without first being converted into control signals expressed in the predetermined format specific to the keyboard.

Abstract: An audio recorder for interfacing with a speech recognition systems includes a microphone having an open front port that senses an audio signal and a closable back port which, when open, senses an audio signal. The microphone also produces an output signal representative of the audio sensed by the microphone. The recorder further includes a mode selector switch having two positions. The first position provides a close-talking mode in which the back port is open so that audio signals are sensed through both the front port and the back port to reduce representation of environmental noise in the output signal. The second position provides a conference-talking mode in which the back port is closed and audio signals are sensed through the front port only and environmental noise is not reduced.

Abstract: A method for editing written text in a text editor which automatically aligns a cursor in the written text on a screen with a particular spoken word during playback of an audio recording. The method may comprise aligning the cursor in a targeted insertion point in response to a user's input, performing one or more editing functions at the targeted insertion point, and realigning the cursor with the spoken words.

Abstract: A method of organizing an acoustic model for speech recognition is comprised of the steps of calculating a measure of acoustic dissimilarity of subphonetic units. A clustering technique is recursively applied to the subphonetic units based on the calculated measure of acoustic dissimilarity to automatically generate a hierarchically arranged model. Each application of the clustering technique produces another level of the hierarchy with the levels progressing from the least specific to the most specific. A technique for adapting the structure and size of a trained acoustic model to an unseen domain using only a small amount of adaptation data is also disclosed.

Abstract: A speech controlled computer user interface communicates between a user and at least one application program. The user interface has a speech layer, an utterance layer, and a discourse layer. The speech layer is in communication with the user and converts between speech messages and text messages. The utterance layer is in communication with the speech layer, and converts between text messages and semantic meaning messages. The discourse layer is in communication with the utterance layer and the at least one application program, and processes messages from the user and the at least one application program, and generates responsive messages to the user and the at least one application program.

Abstract: A modeless large vocabulary continuous speech recognition system is provided that represents an input utterance as a sequence of input vectors. The system includes a common library of acoustic model states for arrangement in sequences that form acoustic models. Each acoustic model is composed of a sequence of segment models and each segment model is composed of a sequence of model states. An input processor compares each vector in a sequence of input vectors to a set of model states in the common library to produce a match score for each model state in the set, reflecting the likelihood that a state is represented by a vector. The system also includes a plurality of recognition modules and associated recognition grammars. The recognition modules operate in parallel and use the match scores with the acoustic models to determine at least one recognition result in each of the recognition modules.

Abstract: Automatic speech recognition word sequence hypotheses are generated using an interleaved forward-backward search. A forward search pass uses relatively simple models for a given block period of time. A backward search pass then goes back over the previous block period using more complex models and the recognition hypotheses generated by the forward search pass. The backward search pass employs a word dependent n-best search having a flat model state organization.

Abstract: A method is given for generating a speaker-dependant model of an utterance that has at least one occurrence. The method includes generating an initial model, having a first resolution, that encodes each of the occurrences of the utterance; and generating at least one additional speaker-specific model, having a different resolution from that of the initial model, of all occurrences of the utterance.

Abstract: A speech recognition system has vocabulary word models having for each word model state both a discrete probability distribution function and a continuous probability distribution function. Word models are initially aligned with an input utterance using the discrete probability distribution functions, and an initial matching performed. From well scoring word models, a ranked scoring of those models is generated using the respective continuous probability distribution functions. After each utterance, preselected continuous probability distribution function parameters are discriminatively adjusted to increase the difference in scoring between the best scoring and the next ranking models.

Abstract: A computer program product residing on a computer readable medium for avoiding headers/footers while synthesizing speech for a reading machine includes instructions for causing the reading machine to determine if text in regions of a document correspond to text of a header or a footer and synthesize speech to read the document aloud to a user of the reading machine, while ignoring those portions of the document that correspond to a header or footer.

Abstract: A method and device for identifying word boundaries in continuous text compares the continuous text to a set of varying length strings to identify candidate word-initial boundaries and candidate word-final boundaries in the continuous text. Each candidate word-initial boundary and candidate word-final boundary has an associated probability value. Each candidate word boundary in the continuous text is identified by calculating a word boundary score for such candidate word boundary using the probability values associated with the candidate word-initial boundaries and candidate word-final boundaries. The set of varying length strings may include words and n-grams.

Abstract: A method for playback of speech in an audio recording. The method comprises performing full word-level recognition of the speech including recognition of silent pauses and filled pauses, suppressing playback of the filled pauses and silent pauses, alerting a listener of the audio recording to locations of suppressed filled pauses and silent pauses during playback of the audio recording, and accepting a user command to disable suppression of any filled pause or silent pause during playback of the audio recording.

Abstract: A system and method of allowing a user to control a computer application with spoken commands, include the steps of processing the spoken commands with a Speech Recognition application into candidate word phrases, and parsing at least one candidate word phrase with a Context Free Grammar (CFG) parser, into a parse tree. A plurality of predefined rewrite rules grouped into a plurality of phases applied are to the parse tree, for rewriting the parse tree. Each of the plurality of rewrite rules includes a pattern matching portion, for matching at least a part of the parse tree, and a rewrite component, for rewriting the matched part. A command string is produced by traversing each terminal node of the modified parse tree. The command string is sent to an interpreter application or directly to the computer application. Possible applications include word processing and other voice-entry systems.

Abstract: A speech recognition system uses language independent acoustic models derived from speech data from multiple languages to represent speech units which are concatenated into words. In addition, the input speech signal which is compared to the language independent acoustic models may be vector quantized according to a codebook which is derived from speech data from multiple languages.

Abstract: A method for having a reading machine spell a word includes retrieving a word to be spelled, displaying the first character of the word on a display while applying a highlighting indicia to the character, synthesizing speech for the character to have the reading machine pronounce the displayed letter of the word, unhighlighting the displayed letter of the word, and retrieving a subsequent letter of the word.

Abstract: A universal voice-data handling system for interfacing between a software application program and at least one speech-related application interface. The voice-data handling system includes a computer system; a run-time module, loaded into and running on the computer system, the run-time module adapted to a specialized task for running concurrently and interactively with the at least one speech-related application interface; and an initialization file, loaded into the computer system, containing parameters specific to the software application program the initialization file being generated by an application, based on the behavior required by the speech-related software application program for specifying a behavior of the run-time module in relation to the software application program the initialization file being generated by an application, based on the behavior required by the speech-related software application program.

Abstract: The present invention describes an apparatus and method for generating phonetico-prosodic parameters of a predetermined message starting from a source message. The predetermined message comprises carriers and phrases. The phonetico-prosodic parameters of the carriers and the phrases are stored in a memory after having been generated off-line. The invention also comprises an apparatus and method for electronically generating a spoken message starting from phonetico-prosodic parameters, stored in said memory. The carriers comprise fixed parts and open slots filled with arguments. The phonetico-prosodic parameters of the arguments to be filled in in the open slots are generated at run time.

Abstract: A method and apparatus for removing convolution noise from a signal such a one carrying speech information. The signal is transformed into a log-spectral domain where a smoothed model is fitted to the log-spectrum subject to constraints of concavity and an overall bandpass shape. The smoothed model has quadratic segments of negative curvature and linear segments, the segments being smoothly joined at breakpoints. The model, which may be recursively updated, is subtracted from each log-spectral data vector.