International Business Machines Corporation

Nuance Communications, Inc.

Shay Ben-David

Ehud Dov Karnin

Amir Geva

Ayal Zaks

Uzi Shvadron

Jacob Sheinvald

David Nahamoo

Charles Willard Cross, Jr

Jeffrey Haskell Derby

Fredy Daniel Neeser

Ron Hoory

Ella Barkan

Zohar Sivan

Jamie H Moreno

Baiju Dhirajlal Mandalia

Thomas Winters Fox

Mattias Marder

Victor V Zyuban

Marc White

Hagai Aronowitz

Alexander Sorin

James Jude Sliwa

Zvi Kons

Jason William Pelecanos

Orith Toledo-Ronen

Evgeny Hazanovich

Zak Mandel

Alexey Roytman

Dorit Naishlos

A method, computer program product, and system for adapting speech recognition of a user's speech is provided. The method includes receiving a first utterance from a user having a duration below a predetermined threshold, identifying at least one further utterance from the user that provides additional information, generating a concatenated utterance by concatenating the first utterance with the at least one further utterance, transmitting the concatenated utterance to a speech recognition server, receiving a transcription of the concatenated utterance from the speech recognition server that includes a transcription of the first utterance, and extracting the transcription of the first utterance from the transcription of the concatenated utterance. The transcription of the first utterance is based on the additional information provided by the at least one further utterance.

Adaptation of speech recognition

Systems and methods for maintaining speaker recognition performance are provided. A method for maintaining speaker recognition performance, comprises training a plurality of models respectively corresponding to speaker recognition scores from a plurality of speakers over a plurality of sessions, and using the plurality of models to conclude whether a speaker seeking access to an environment is a non-ideal target speaker or a non-ideal non-target speaker. Using the plurality of models to conclude comprises calculating a first probability that the speaker seeking access is the non-ideal target speaker, calculating a second probability that the speaker seeking access is the non-ideal non-target speaker, and determining whether the first probability, the second probability or a sum of the first probability and the second probability is above a probability threshold.

Detection of target and non-target users using multi-session information

Synchronizing a data stream with an associated metadata stream by receiving a data stream and a metadata stream having a plurality of metadata events associated with the data stream, identifying within the data stream a plurality of data events, matching each of the data events to one of the metadata events in accordance with a matching criterion, and synchronizing the data stream with the metadata stream by effecting a relative time shift between the metadata stream and the data stream in accordance with a time shift adjustment value that results in the smallest sum of absolute differences between time indices of each matched data event and metadata event.

Synchronization of data streams with associated metadata streams using smallest sum of absolute differences between time indices of data events and metadata events

Methods, apparatus, and computer program products are disclosed for synchronizing distributed speech recognition ('DSR') that include receiving in a DSR client notification from a voice server of readiness to conduct speech recognition and, responsive to the receiving, transmitting by the DSR client, from the DSR client to the voice server, speech for recognition.

Synchronizing distributed speech recognition

Method, system, and computer program product for voice transformation are provided. The method includes transforming a source speech using transformation parameters, and encoding information on the transformation parameters in an output speech using steganography, wherein the source speech can be reconstructed using the output speech and the information on the transformation parameters. A method for reconstructing voice transformation is also provided including: receiving an output speech of a voice transformation system wherein the output speech is transformed speech which has encoded information on the transformation parameters using steganography; extracting the information on the transformation parameters; and carrying out an inverse transformation of the output speech to obtain an approximation of an original source speech.

Voice transformation with encoded information

Systems and methods for detecting end of a transaction in a computing environment are provided. The method comprises determining a target area in a graphical user environment displayed on a display screen, wherein a change is expected to occur when end of a transaction is reached; masking the target area at least partially to remove content included in the target area that is present before or after the transaction was initiated; monitoring the target area for change in content; and detecting the end of the transaction when the content of the target area has changed.

Recognition techniques to enhance automation in a computing environment

A voice processing system includes a real-time voice server, which is arranged to process real-time voice processing tasks for clients of the system. A gateway processor is arranged to accept from a client a request to perform an off-line voice processing task, to convert the off-line voice processing task into an equivalent real-time voice processing task, to invoke the voice server to process the equivalent real-time voice processing task, and to output a result of the equivalent real-time voice processing task.

Distributed off-line voice services

Methods, apparatus, and computer program products are described for document session replay for multimodal applications. including identifying, by a multimodal browser in dependence upon a log produced by a Form Interpretation Algorithm ('FIA') during a previous document session with a user, a speech prompt provided by a multimodal application in the previous document session; identifying, by a multimodal browser in replay mode in dependence upon the log, a response to the prompt provided by a user of the multimodal application in the previous document session; retrieving, by the multimodal browser in dependence upon the log, an X+V page of the multimodal application associated with the speech prompt and the response; rendering, by the multimodal browser, the visual elements of the retrieved X+V page; replaying, by the multimodal browser, the speech prompt; and replaying, by a multimodal browser, the response.

Document session replay for multimodal applications

Methods and systems are provided for remote tuning and debugging of an automatic speech recognition system. Trace files are generated on-site from input speech by efficient, lossless compression of MFCC data, which is merged with compressed pitch and voicing information and stored as trace files. The trace files are transferred to a remote site where human-intelligible speech is reconstructed and analyzed. Based on the analysis, parameters of the automatic speech recognition system are remotely adjusted.

Remote tracing and debugging of automatic speech recognition servers by speech reconstruction from cepstra and pitch information

A computer-implemented method for processing paper forms includes capturing at a computer system an image of a paper form in which information has been filled-in. A location identifier is extracted from the image. The location identifier indicates an address in a storage location external to the computer system, at which the filled-in information is electronically stored. The information is retrieved responsively to the location identifier by communication with the storage location via a wide area network (WAN), so as to convey the information electronically from the storage location to the computer system. The information is processed using a data processing application running on the computer system.

Automated processing of paper forms using remotely-stored form content

A method for channel estimation, includes receiving an input signal due to a superposition of waveforms encoding symbols transmitted respectively by a plurality of transmitters in a common frequency band. Each transmitter among the plurality of transmitters modulates the symbols with a respective spreading code. The input signal is sampled at sampling intervals over the duration of an observation period, thus providing a sequence of samples. The sequence of complex samples is processed to estimate a channel response for each of the transmitters by maximizing, individually for each of the transmitters and based on the respective spreading code of the transmitter, a likelihood of correctly decoding the transmitted symbols using the estimated channel response.

Code-division-multiple-access (DS-CDMA) channels with aperiodic codes

A method for determining vectorization configurations in a computer processor architecture, the method including identifying a vectorizable loop in a computer program, identifying a memory access pattern of data required for implementing the loop in the architecture, computing a set of candidate configurations of resources required for vectorizing the data in the architecture, where the computing step includes configuring a vector pointer register of the architecture in support of either of reorder-on-read use and reorder-on-write use of a vector element file of the architecture, selecting one of the candidates in accordance with predefined selection criteria, and implementing the selected vectorization configuration in the architecture.

Vectorization in a SIMdD DSP architecture

A method for multi-user detection includes receiving a complex input signal due to a superposition of waveforms encoding symbols in a constellation of fixed magnitude and variable phase, which symbols are transmitted respectively by a plurality of transmitters in a common frequency band. The complex input signal is sampled at sampling intervals over the duration of an observation period to provide a sequence of complex samples. The sequence of complex samples is processed to determine soft decision values corresponding to the symbols transmitted by the plurality of the transmitters in the observation period, while constraining the soft decision values to a circle in a complex plane. The soft decision values are projected onto the constellation to estimate the transmitted symbols.

Enhancing CDMA multiuser detection by constraining soft decisions

A method for multi-user detection includes receiving a complex input signal due to a superposition of waveforms encoding symbols in a real-valued constellation, which are transmitted respectively by a plurality of transmitters in a common frequency band. The complex input signal is sampled at sampling intervals over the duration of an observation period to provide a sequence of complex samples. The sequence of complex samples is processed to determine soft decision values corresponding to the symbols transmitted by the plurality of the transmitters in the observation period, while constraining the soft decision values to be real values. The soft decision values are then projected onto the constellation to estimate the transmitted symbols.

Growing community of inventors

Haifa, Israel

Shay Ben-David