Abstract: Systems, devices, media, and methods are presented for generating a language detection model of a language analysis system. The systems and methods access a set of messages including text elements and convert the set of messages into a set of training messages. The set of training messages are configured for training a language detection model. The systems and methods train a classifier based on the set of training messages. The classifier has a set of features representing word frequency, character frequency, and a character ratio. The systems and methods generate a language detection model based on the classifier and the set of features.

Abstract: In one embodiment, a method includes receiving a first user request to execute a first task comprising one or more sub-tasks from a client system associated with a user, receiving a second user request to suspend the first task from the client system, storing a dialog state for the first task in response to the second user request, wherein the dialog state comprises results of executed sub-tasks and a status indicating the first task is suspended, receiving a third user request to resume the first task, accessing the dialog state for the first task, and providing a prompt to resume the first task at the client system.

Abstract: Methods and systems for generation of a database schema compliant search query based on a natural language input are described herein. Natural language input may be received from a computing device. The natural language input may be associated with multiple search requests to a database. The natural language input may be parsed into a plurality of segments. The plurality of segments may be, for example, one or more words of a text string. At least one identifier for the plurality of segments may be associated with one or more confidence values. The natural language input may be converted into a single search query based on the confidence values and/or on a set of rules. The single search query may be initiated with respect to the database. The single search query may fetch content more efficiently than the multiple search requests.

Abstract: A method, an apparatus, a device and a storage medium for outputting information are provided. The method includes: acquiring a to-be-translated sentence; determining a foreign language sentence matching the to-be-translated sentence from a preset set of sentence pairs, where the set of sentence pairs includes local language sentences and corresponding foreign language sentences; determining a target foreign language sentence of the to-be-translated sentence according to the to-be-translated sentence and the determined foreign language sentence; and outputting the target foreign language sentence.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for coding speech using neural networks. One of the methods includes obtaining a bitstream of parametric coder parameters characterizing spoken speech; generating, from the parametric coder parameters, a conditioning sequence; generating a reconstruction of the spoken speech that includes a respective speech sample at each of a plurality of decoder time steps, comprising, at each decoder time step: processing a current reconstruction sequence using an auto-regressive generative neural network, wherein the auto-regressive generative neural network is configured to process the current reconstruction to compute a score distribution over possible speech sample values, and wherein the processing comprises conditioning the auto-regressive generative neural network on at least a portion of the conditioning sequence; and sampling a speech sample from the possible speech sample values.

Abstract: A speech interface device is configured to receive response data from a remote speech processing system for responding to user speech. This response data may be enhanced with information such as remote NLU data. The response data from the remote speech processing system may be compared to local NLU data to improve a speech processing model on the device. Thus, the device may perform supervised on-device learning based on the remote NLU data. The device may determine differences between the updated speech processing model and an original speech processing model received from the remote system and may send data indicating these differences to the remote system. The remote system may aggregate data received from a plurality of devices and may generate an improved speech processing model.

Abstract: A method for training parameters of a first domain adaptation model. The method includes evaluating a cycle consistency objective using a first task specific model associated with a first domain and a second task specific model associated with a second domain, and evaluating one or more first discriminator models to generate a first discriminator objective using the second task specific model. The one or more first discriminator models include a plurality of discriminators corresponding to a plurality of bands that corresponds domain variable ranges of the first and second domains respectively. The method further includes updating, based on the cycle consistency objective and the first discriminator objective, one or more parameters of the first domain adaptation model for adapting representations from the first domain to the second domain.

Abstract: Methods and systems are provided for enhanced audio experiences in VR/AR applications. The apparatuses of this disclosure are adapted to record multiple binaural stereo pairs and play back select binaural pairs corresponding to users' head positions. A substantially spherical microarray is utilized in various embodiments for recording multiple binaural stereo pairs. A VR/AR headset is further adapted to track a user's head positions and dynamically play back binaural sound pairs corresponding to the head positions.

Abstract: A method, a computer-readable medium, and an apparatus for feature learning are provided. The apparatus may receive a data sample as an input to a feature learning model. The apparatus may calculate a reconstruction error based on the data sample and a plurality of features of the feature learning model. The apparatus may determine whether the reconstruction error satisfies a first threshold. The apparatus may add a feature into the feature learning model to represent the data sample if the data sample satisfies the first threshold. The apparatus may determine whether the reconstruction error satisfies a second threshold. The apparatus may ignore the data sample if the reconstruction error satisfies the second threshold. The apparatus may update the weights associated with the plurality of features of the feature learning model if the reconstruction error satisfies neither the first threshold nor the second threshold.

Abstract: The disclosed subject matter relates to a system and method for providing an automated assistant that retrieves information from a knowledge base in response to a user's natural language question. A user's natural language question voice is transformed into an intermediate representation. From the intermediate representation, a cypher query is generated which may be used to query the database. The query results are provided in response to the user. The transformation into the intermediate representation is database independent while the cypher query is dependent upon the database queried.

Abstract: The present disclosure relates generally to techniques for analyzing and improving a bot system, and more particularly to an analytic system integrated with a bot system for monitoring, analyzing, visualizing, diagnosing, and improving the performance of the bot system. For example, an analytic system is integrated with a bot system for monitoring, analyzing, visualizing, and improving the performance of the bot system. The analytic system monitors events occurred in conversations between end users and the bot system, aggregates and analyzes the collected events, and provides information regarding the conversations graphically on a graphic user interface as insights reports at different generalization levels. The insights reports offer developer-oriented analytics to pinpoint issues with skills so a user can address them before they cause problems.

Abstract: A method for separating and authenticating speech of a speaker on an audio stream of speakers over an audio channel may include receiving audio stream data of the audio stream with speech from a speaker to be authenticated speaking with a second speaker. A voiceprint may be generated for each data chunk in the audio stream data divided into a plurality of data chunks. The voiceprint for each data chunk may be assessed as to whether the voiceprint has speech belonging to the speaker to be authenticated or to the second speaker using representative voiceprints of both speakers. An accumulated voiceprint may be generated using the verified data chunks with speech of the speaker to be authenticated. The accumulated voiceprint may be compared to the reference voiceprint of the speaker to be authenticated for authenticating the speaker speaking with the second speaker over the audio channel.

Abstract: Techniques for determining an intent for a user input in a dialog are described. The system processes historic interaction data that is structured based skills and intents, with each skill-intent pair being associated with one or more past user inputs received by the system, one or more sample inputs, and one or more alternative representations of the user inputs. Based on processing of the historic interaction data and dialog data of previous turns of the dialog, the system determines potential intents for the user input of the current turn of the dialog. The potential intents may correspond to a presently active skill or another skill, enabling the user to interact with another skill during the dialog.

Abstract: A method of attention-based end-to-end (A-E2E) automatic speech recognition (ASR) training, includes performing cross-entropy training of a model, based on one or more input features of a speech signal, determining a posterior probability vector at a time of a first wrong token among one or more output tokens of the model of which the cross-entropy training is performed, and determining a loss of the first wrong token at the time, based on the determined posterior probability vector. The method further includes determining a total loss of a training set of the model of which the cross-entropy training is performed, based on the determined loss of the first wrong token, and updating the model of which the cross-entropy training is performed, based on the determined total loss of the training set.

Abstract: Embodiments described herein provide a composed variational natural language generation (CLANG) model that is configured to generate training samples for few-shot intents. Specifically, the CLANG model may build connections between existing training samples of many-shot intents and new training samples of few-shot intents by modeling an intent as a combination of a domain and an action. In this way, the CLANG model transfers knowledge from existing many-shot intents to few-shot intents in natural language generation by learning how to compose utterances with many-shot intents and transferring such knowledge to few-shot intents.

Abstract: Systems and methods for determining to perform an action of a query using a trained natural language model of a natural language understanding (NLU) system are disclosed herein. A text string corresponding to a prescribed action includes at least a content entity is received. A determination is made as to whether the text string corresponds to an audio input of a first group. In response to determining the text string corresponds to an audio input of a first group, a determination is made as to whether the text string includes an obsequious expression. In response to determining the text string corresponds to an audio input of a first group and in response to determining the text string includes an obsequious expression, a determination is made to perform the prescribed action.

Abstract: Aspects of the disclosure relate to enhanced dynamic training response output generation control systems with enhanced dynamic training response output determinations. A computing platform may receive, from the user device and in response to an initial dynamic training interface, a training request input. The computing platform may send, to an NLU engine, the training request input and commands directing the NLU engine to perform natural language understanding and processing on the training request input to determine a natural language result output. Using the natural language result output, the computing platform may determine third party data sources that correspond to the natural language result output, and may request source data from the third party data sources. Using the source data and the natural language result output, the computing platform may generate a dynamic training response output, and may direct the user device to cause display of the dynamic training response output.

Abstract: A signal processing apparatus includes one or more processors. The one or more processors perform first-type signal processing on an input signal using a neural network, and output a first-type output signal. The one or more processors convert the first-type output signal into a second-type output signal for calculating a first-type loss related to accuracy of second-type signal processing performed by another signal processing device The one or more processors calculate the first-type loss based on the second-type output signal and a correct signal. The one or more processors optimize parameters of the neural network based on the first-type loss.

Abstract: The exemplary embodiments disclose a system and method, a computer program product, and a computer system for resolving the intents of user expression. The exemplary embodiments may include receiving a user expression, receiving a first resolver having an input class and an output class based on the user expression, determining whether the first resolver can be resolved based on the user expression, and based on determining that the first resolver can be resolved based on the user expression, resolving the first resolver.

Abstract: A method for execution by a computing device includes obtaining a phrase that includes string of words and generating a valid sequence of words utilizing the phrase. The method further includes identifying a set of identigens for each word of the valid sequence of words to produce sets of identigens. The method further includes identifying, for each identigen of the sets of identigens, a word type associated with phrase structure grammar rules to produce sets of identigen-type associations. The method further includes interpreting, utilizing the phrase structure grammar rules, the sets of identigen-type associations to produce an entigen group. The entigen group represents a most likely interpretation of the phrase.