Abstract: Techniques for medical coding include applying a natural language understanding (NLU) engine to a free-form text documenting a clinical patient encounter, to derive a first set of one or more medical billing codes for the clinical patient encounter and a link between each code in the first set and a corresponding portion of the free-form text. The first set of codes may be compared to a second set of one or more medical billing codes approved by one or more human users for the patient encounter, to identify at least one code in the first set that overlaps with at least one code in the second set. The code in the second set approved by the one or more human users may be retained instead of the overlapping code in the first set derived by the NLU engine.

Abstract: Techniques for question answering involve receiving, from a user, a text input expressing a question in natural language. The text input may be analyzed, including identifying in the question at least one first portion answerable from at least one structured data source, and at least one second portion answerable from at least one unstructured data source. At least one first query configured for the structured data source(s) may be constructed from the at least one first portion of the question and applied to the structured data source(s) to retrieve first answer information for the at least one first portion of the question. At least one second query configured for the unstructured data source(s) may be constructed from the at least one second portion of the question and applied to the unstructured data source(s) to retrieve second answer information for the at least one second portion of the question.

Abstract: Techniques for determining a clinician's intent to order an item may include processing a free-form narration, of an encounter with a patient, narrated by a clinician, using a natural language understanding engine implemented by one or more processors, to extract at least one clinical fact corresponding to a mention of an orderable item from the free-form narration. The processing may comprise distinguishing between whether the at least one clinical fact indicates an intent to order the orderable item or does not indicate an intent to order the orderable item. In response to determining that the at least one clinical fact indicates an intent to order the orderable item, an order may be generated for the orderable item.

Abstract: An assignment device (1) assigns word class information (WKI) to one or more words of text information (ETI). Based on word-class sequence information (WK-AI) formed from this assigned word class information (WKI), actions (A) are executed in order to notify the user of conflicts or to provide the user with background information (HI) relating to words in the text information (TT).

Abstract: An arrangement and corresponding method are described for multi-domain natural language processing. Multiple parallel domain pipelines are used for processing a natural language input. Each domain pipeline represents a different specific subject domain of related concepts. Each domain pipeline includes a mention module that processes the natural language input using natural language understanding (NLU) to determine a corresponding list of mentions, and an interpretation generator that receives the list of mentions and produces a rank-ordered domain output set of sentence-level interpretation candidates. A global evidence ranker receives the domain output sets from the domain pipelines and produces an overall rank-ordered final output set of sentence-level interpretations.

Abstract: A method of detecting an occurrence of splicing in a speech signal includes comparing one or more discontinuities in the test speech signal to one or more reference speech signals corresponding to the test speech signal. The method may further include calculating a frame-based spectral-like representation ST of the speech signal, and calculating a frame-based spectral-like representation SE of a reference speech signal corresponding to the speech signal. The method further includes aligning ST and SE in time and frequency, calculating a distance function associated with aligned ST and SE, and evaluating the distance function to determine a score. The method also includes comparing the score to a threshold to detect if splicing occurs in the speech signal.

Abstract: Some embodiments provide techniques performed by at least one voice agent. The techniques include receiving voice input; identifying at least one application program as relating to the received voice input; and displaying at least one selectable visual representation that, when selected, causes focus of the computing device to be directed to the at least one application program identified as relating to the received voice input.

Abstract: A method, computer program product, and computer system for addressing acoustic signal reverberation is provided. Embodiments may include receiving, at one or more microphones, a first audio signal and a reverberation audio signal. Embodiments may further include processing at least one of the first audio signal and the reverberation audio signal. Embodiments may also include limiting a model based reverberation equalizer using a temporal constraint for direct sound distortions, the model based reverberation equalizer configured to generate one or more outputs, based upon, at least in part, at least one of the first audio signal and the reverberation audio signal.

Abstract: Methods and apparatus for communicating between virtual agents associated with users of electronic devices connected via at least one network. A first user may instruct an associated first virtual agent to invoke a communication session with a second virtual agent associated with a second user. To invoke the communication session, the first virtual agent may send an outgoing communication to the second virtual agent and the outgoing communication may instruct the second virtual agent to perform at least one action on behalf of the first user. Virtual agents associated with different users may alternatively communicate with each other in the absence of user interaction to perform a collaborative action.

Abstract: Techniques for performing multi-style speech synthesis. The techniques include using at least one computer hardware processor to perform: obtaining input comprising text and an identification of a desired speaking style to use in rendering the text as speech; identifying a plurality of speech segments for use in rendering the text as speech, the identifying comprising identifying a first speech segment recorded and/or synthesized in a first speaking style that is different from the desired speaking style based at least in part on a measure of similarity between the desired speaking style and the first speaking style; synthesizing speech from the text in the desired speaking style at least in part by using the first speech segment; and outputting the synthesized speech.

Abstract: Embodiments of the present disclosure may include a system and method for speech enhancement using the coherent to diffuse sound ratio. Embodiments may include receiving an audio signal at one or more microphones and controlling one or more adaptive filters of a beamformer using a coherent to diffuse ratio (“CDR”).

Abstract: The present disclosure is directed towards an audio conferencing method. Some embodiments may include receiving, at a first mixing device, a first audio stream from one or more participant conferencing devices. The method may further include generating a top-N voice stream at the first mixing device, wherein the top-N voice stream corresponds with at least one top-N talker and wherein the identification of the at least one top-N talker is based upon, at least in part, an activity ranking. The method may also include receiving the top-N voice stream at a centralized mixing device and generating at least one mixed audio stream at the centralized mixing device.

Abstract: Inferring a natural language grammar is based on providing natural language understanding (NLU) data with concept annotations according to an application ontology characterizing a relationship structure between application-related concepts for a given NLU application. An application grammar is then inferred from the concept annotations and the application ontology.

Abstract: Designing a natural language understanding (NLU) model for an application from scratch can be difficult for non-experts. A system can simplify the design process by providing an interface allowing a designer to input example usage sentences and build an NLU model based on presented matches to those example sentences. In one embodiment, a method for initializing a workspace for building an NLU system includes parsing a sample sentence to select at least one candidate stub grammar from among multiple candidate stub grammars. The method can include presenting, to a user, respective representations of the candidate stub grammars selected by the parsing of the sample sentence. The method can include enabling the user to choose one of the respective representations of the candidate stub grammars. The method can include adding to the workspace a stub grammar corresponding to the representation of the candidate stub grammar chosen by the user.

Abstract: Methods and apparatus to process microphone signals by a speech enhancement module to generate an audio stream signal including first and second metadata for use by a speech recognition module. In an embodiment, speech recognition is performed using endpointing information including transitioning from a silence state to a maybe speech state, in which data is buffered, based on the first metadata and transitioning to a speech state, in which speech recognition is performed, based upon the second metadata.

Abstract: A mobile device is adapted for automatic speech recognition (ASR). A user interface for interaction with a user includes an input microphone for obtaining speech inputs from the user for automatic speech recognition, and an output interface for system output to the user based on ASR results that correspond to the speech input. A local controller obtains a sample of non-ASR audio from the input microphone for ASR-adaptation to channel-specific ASR characteristics, and then provides a representation of the non-ASR audio to a remote ASR server for server-side adaptation to the channel-specific ASR characteristics, and then provides a representation of an unknown ASR speech input from the input microphone to the remote ASR server for determining ASR results corresponding to the unknown ASR speech input, and then provides the system output to the output interface.

Abstract: According to some aspects, a system for detecting a designated wake-up word is provided, the system comprising a plurality of microphones to detect acoustic information from a physical space having a plurality of acoustic zones, at least one processor configured to receive a first acoustic signal representing the acoustic information received by the plurality of microphones, process the first acoustic signal to identify content of the first acoustic signal originating from each of the plurality of acoustic zones, provide a plurality of second acoustic signals, each of the plurality of second acoustic signals substantially corresponding to the content identified as originating from a respective one of the plurality of acoustic zones, and performing automatic speech recognition on each of the plurality of second acoustic signals to determine whether the designated wake-up word was spoken.

Abstract: A method of providing a task assistant is described. The task assistant is designed to receive input from a user through multimodal input including a plurality of speech input, typing input, and touch input, determine the meaning of the input, and determining whether there is a context based on prior interactions with the user. The method further to generate an interpreted input based on a combination of the input and the context, and providing a formatted query to an application. The method further to receive data from the application in response to the formatted query, and provide a response to the user through multimodal output including a plurality of: speech output, text output, non-speech audio output, haptic output, and visual non-text output. The method further to update the context based on the interpreted input.

Abstract: Methods and apparatus to process a user input on independent applications that provide classifier outputs to an arbitration module, which selects one of the application to respond to the user input. The classifier outputs include a probability that the user input is in domain for the application functionality.

Abstract: Systems, methods, and computer-readable storage devices for performing per-channel automatic speech recognition. An example system configured to practice the method combines a first audio signal of a first speaker in a communication session and a second audio signal from a second speaker in the communication session as a first audio channel and a second audio channel. The system can recognize speech in the first audio channel of the recording using a first model specific to the first speaker, and recognize speech in the second audio channel of the recording using a second model specific to the second speaker, wherein the first model is different from the second model. The system can generate recognized speech as an output from the communication session. The system can identify the models based on identifiers of the speakers, such as a telephone number, an IP address, a customer number, or account number.