Abstract: Some embodiments provide a method of training a MT network to detect a wake expression that directs a digital assistant to perform an operation based on a request that follows the expression. The MT network includes processing nodes with configurable parameters. The method iteratively selects different sets of input values with known sets of output values. Each of a first group of input value sets includes a vocative use of the expression. Each of a second group of input value sets includes a non-vocative use of the expression. For each set of input values, the method uses the MT network to process the input set to produce an output value set and computes an error value that expresses an error between the produced output value set and the known output value set. Based on the error values, the method adjusts configurable parameters of the processing nodes of the MT network.

Abstract: This disclosure relates to method and system for annotating tokens for natural language processing (NLP). In one embodiment, the method may include segmenting a plurality of corpus based on each of a plurality of instances, deriving a plurality of entities for each of the plurality of instances based on at least one of a machine learning technique or a deep learning technique, determining a word vector for each of the plurality of entities associated with each of the plurality of instances, and labelling a plurality of tokens for each of the plurality of instances. It should be noted that the plurality of tokens associated with the plurality of entities may be identified based on a frequency of each of the plurality of entities.

Abstract: A voice-interaction device includes a plurality of input and output components configured to facilitate interaction between the voice-interaction device and a target user. The plurality of input and output components may include a microphone configured to sense sound and generate an audio input signal, a speaker configured to output an audio signal to the target user, and an input component configured to sense at least one non-audible interaction from the target user. A context controller monitors the plurality of input and output components and determines a current use context. A virtual assistant module facilitates voice communications between the voice-interaction device and the target user and configures one or more of the input and output components in response to the current use context. The current use context may include whisper detection, target user proximity, gaze direction tracking and other use contexts.

Abstract: Techniques are provided for building a dialog-state specific contextual language understanding system using subsumption logic. Information establishing conversational rules identifying the conversational dialog is received to present in respective dialog states. Each rule has a Boolean trigger expression of predicates for testing the conversational state together with logical connectives to identify when the rule is applicable. Subsumption logic is used to arrange the rules into a directed acyclic graph (DAG) where more specific rules are preferred to more general rules. During a conversation, the DAG is used to filter the triggered rules to only the most specific triggered rules from which a rule to run is selected. This structure makes it easier to build conversational systems because rules can be added or removed without having to change or reason over other rules. The rules also act as a constraint to help machine learned selection systems converge with less data.

Abstract: Methods, systems, and computer-readable media for artificially generating a revoiced media stream and maintaining original volume changes of a character in the revoiced media stream are provided. For example, a media stream including an individual speaking may be obtained. A transcript of the media stream may be obtained. The transcript of the media stream may be translated to a target language. A revoiced media stream in which the translated transcript in the target language is spoken by a virtual entity may be generated, wherein a ratio of the volume levels between first and second sets of words in the revoiced media stream is substantially identical to the ratio of volume levels between corresponding first and second utterances in the received media stream.

Abstract: A dialogue system including a processor, a memory, an audio input apparatus, an audio output apparatus, a touch input apparatus, and a display unit. The processor receives an input from the voice or a touch input apparatus, analyzes the content of the input, and selects a scenario corresponding to the input data from preset scenario information. The processor generates the output data specified in the scenario, calculates the priority of the input data, and determines the presence or absence of the scenario being prepared for the output data. When there is the scenario under generating the output data, the processor changes the output method of the scenario to be executed based on the priority.

Abstract: A method includes generating first audio data based on sound detected by a sound sensor at a first time. The method further includes transmitting, via one or more communication interfaces, the first audio data to another device during a communication session based on a determination that the sound sensor is unmuted with respect to the communication session at the first time. The method further includes generating second audio data based on sound detected by the sound sensor at a second time. The method further includes refraining from transmitting the second audio data to the other device during the communication session based on a determination that the sound sensor is muted with respect to the communication session at the second time. The method further includes initiating a natural language processing operation on the second audio data based on detecting a wake phrase in the second audio data.

Abstract: A natural language processing system includes logic circuitry that receives, from a first user channel, a user query including natural language (NL) input in a colloquial format from an end user, selects a first NLP application from a plurality of NLP applications, transforms, using the first NLP application, the NL input into a plurality of language tokens, extracts a command trigger from the language tokens, determines that the command trigger is linked to a command executable by a first wagering game channel, transmits a command query incorporating the command to the first wagering game channel, receives a command reply including wagering game data associated with the command from the first wagering game channel, generates, using the first NLP application, an NL response that includes the wagering game data and is structured in a colloquial format, and transmits the NL response to the end user via the first user channel.

Abstract: Systems and methods of real-time vehicle-based analytics are provided herein. An example method includes collecting at least one of images, video, or audio of a user when operating a vehicle; analyzing the at least one of the images, video, or audio to determine an emotion or sentiment of the user when interacting with one or more features of the vehicle; identifying user actions that precede a point in time where the emotion or sentiment of the user was detected, wherein the user actions relate to the one or more vehicle features; classifying at least one of the user actions and the one or more vehicle features with the emotion or sentiment; and storing the user actions, the one or more vehicle features, and the emotion or sentiment.

Abstract: Background noise estimators and methods are disclosed for estimating background noise in an audio signal. Some methods include obtaining at least one parameter associated with an audio signal segment, such as a frame or part of a frame, based on a first linear prediction gain, calculated as a quotient between a residual signal from a 0th-order linear prediction and a residual signal from a 2nd-order linear prediction for the audio signal segment. A second linear prediction gain is calculated as a quotient between a residual signal from a 2nd-order linear prediction and a residual signal from a 16th-order linear prediction for the audio signal segment. Whether the audio signal segment comprises a pause is determined based at least on the obtained at least one parameter; and a background noise estimate is updated based on the audio signal segment when the audio signal segment comprises a pause.

Abstract: Negation scope analysis for negation detection is provided. In various embodiments, a phrase is read from a report collection. The phrase is searched for at least one of a predetermined set of negation keywords. A dependency parse tree is generated of the phrase. The dependency parse tree is traversed starting with the at least one of the predetermined set of negation keywords. Based on the traversal, a plurality of words of the phrase are determined that are spanned by the at least one of the predetermined set of negation keywords.

Abstract: The disclosed computer-implemented method for performing natural language translation in AR may include accessing an audio input stream that includes words spoken by a speaking user in a first language. The method may next include performing active noise cancellation on the words in the audio input stream so that the spoken words are suppressed before reaching a listening user. Still further, the method may include processing the audio input stream to identify the words spoken by the speaking user, and translating the identified words spoken by the speaking user into a second, different language. The method may also include generating spoken words in the second, different language using the translated words, and replaying the generated spoken words in the second language to the listening user. Various other methods, systems, and computer-readable media are also disclosed.

Abstract: A method, system, and computer program product for processing an encoded audio signal is described. In one exemplary embodiment, the system receives an encoded low-frequency range signal and encoded energy information used to frequency shift the encoded low-frequency range signal. The low-frequency range signal is decoded and an energy depression of the decoded signal is smoothed. The smoothed low-frequency range signal is frequency shifted to generate a high-frequency range signal. The low-frequency range signal and high-frequency range signal are then combined and outputted.

Abstract: A speech recognition method and apparatus for performing speech recognition in response to an activation word determined based on a situation are provided. The speech recognition method and apparatus include an artificial intelligence (AI) system and its application, which simulates functions such as recognition and judgment of a human brain using a machine learning algorithm such as deep learning.

Abstract: A method and device are provided for translating object information and acquiring derivative information, including obtaining, based on the acquired source-object information, target-object information corresponding to the source object by translation, and outputting the target-object information. A language environment corresponding to the source object is different from a language environment corresponding to the target object. By applying the present disclosure, the range of machine translation subjects can be expanded, and the applicability of translation can be enhanced, a user's requirements on translation of objects can be met.

Abstract: A time scaler for providing a time scaled version of an input audio signal is configured to compute or estimate a quality of a time scaled version of the input audio signal obtainable by a time scaling of the input audio signal. The time scaler is configured to perform the time scaling of the input audio signal in dependence on the computation or estimation of the quality of the time scaled version of the input audio signal obtainable by the time scaling. An audio decoder has such a time scaler.

Abstract: A method and an apparatus for detecting correctness of a pitch period, where the method for detecting correctness of a pitch period includes determining, according to an initial pitch period of an input signal in a time domain, a pitch frequency bin of the input signal, where the initial pitch period is obtained by performing open-loop detection on the input signal, determining, based on an amplitude spectrum of the input signal in a frequency domain, a pitch period correctness decision parameter, associated with the pitch frequency bin, of the input signal, and determining correctness of the initial pitch period according to the pitch period correctness decision parameter. Hence, the method and apparatus for detecting correctness of the pitch period improve, based on a relatively less complex algorithm, accuracy of detecting correctness of the pitch period.

Abstract: A method is provided in accordance with an aspect of the present disclosure. The method includes detecting at least one voice input from a user of an electronic device, transforming the at least one voice input into a text structure including at least one word, and determining a current context scope of the electronic device. The method also includes comparing the text structure to a plurality of existing text structures, where each of the existing text structure associated with a command for an action on the electronic device. The method further includes identifying, when the text structure matches with at least one of the existing text structures, a command to correspond to the at least one voice input from the user, and performing an action on the electronic device based on the identified command.

Abstract: Systems, methods, and computer-readable storage devices for crowd-sourced data labeling. The system requests a respective response from each of a set of entities. The set of entities includes crowd workers. Next, the system incrementally receives a number of responses from the set of entities until one of an accuracy threshold is reached and m responses are received, wherein the accuracy threshold is based on characteristics of the number of responses. Finally, the system generates an output response based on the number of responses.

Abstract: A method includes receiving data in a first series of blocks each having a first number of audio samples and repackaging the data into a second series of blocks each having a second number of audio samples. The second number of audio samples is a non-integer fraction of the first number of audio samples. The method further includes transmitting the second series of blocks over a series of fixed duration time intervals, and adjusting the payload of adjacent time intervals to reduce jitter in the transmission of the second series of blocks.