Abstract: A method for pre-training a language model includes: constructing a pre-training language data set, in which the pre-training language data set comprises unsupervised language data and supervised language data; generating a hierarchical multi-template and multi-task language data set based on the pre-training language data set; and pre-training the language model based on the hierarchical multi-template and multi-task language data set.

Abstract: An example system for performing call center tasks with the assistance of a human operator comprises a plurality of remote electronic devices and a system server. Each of the plurality of remote electronic devices has a speaker, a microphone and a screen. The system server has a data store configured to store personal information and a dialing platform configured to connect the remote electronic devices with customer devices. The system server may transmit fragments of personal information to the remote electronic device and the remote electronic device may display the fragments of personal information on its screen as the human operator engages in the call with the customer over the remote electronic device.

Abstract: Disclosed in some examples are methods, systems, and machine readable mediums which provide summaries of topics determined within a corpus of documents. These summaries may be used by customer service associates, analysts, or other users to quickly determine both topics discussed and contexts of those topics over a large corpus of text. For example, a corpus of documents may be related to customer complaints and the topics may be summarized to produce summaries such as “credit report update due to stolen identity.” These summarizations may be used to efficiently spot trends and issues.

Abstract: Method and server for performing domain-specific translation of sentences. The method includes generating, by the server, an augmented sequence of input tokens based on an input sentence in the first language and a hint inserted into the input sentence. The method includes iteratively generating, using the NN, a sequence of output tokens based on the augmented sequence of input tokens. The method includes generating a second sentence in the second language using the sequence of output tokens, the second sentence including the domain-specific translation of the given input word.

Abstract: The invention relates to computer-implemented systems and methods for analyzing medical images and records in various formats and structures from disparate sources and providers to provide answers to specific medical-related questions. An embodiment of the present invention is directed to accessing a health care form, identifying related text, leveraging computer vision technology to group nearby text, and applying deep learning models to identify fields from the text. This information may then be used to create a structured standardized output.

Abstract: Implementations relate to an automated assistant that can determine whether to respond to inputs in an environment according to whether radar data indicates a user is present. When user presence is detected, the automated assistant can virtually segment the environment and apply certain operational parameters to certain segments of the environment. For instance, the automated assistant can enable an input detection feature, such as warm word detection, for a segmented portion of the environment in which a user is detected. In this way, false positives can be mitigated for instances in which environmental and/or user sounds are detected by the automated assistant but do not originate from a particular segment of the environment. Other parameters, such as varying confidence thresholds and/or speech processing biasing, can be temporarily enforced for different segments of an environment in which a user is detected.

Abstract: A device may include a processor configured to obtain a time series of documents; determine query scores for particular documents of the time series of documents based on a set of query terms; and determine time scores for the particular documents based on timestamps associated with the particular documents. The processor may be further configured to compute document relevancy scores for the particular documents based on a combination of the query scores and the time scores for the particular documents; order the particular documents based on the computed relevancy scores; and generate a document summary by applying a document summarizer applied to the ordered particular documents.

Abstract: An audio output device may include one or more audio sensors configured to detect ambient sound and noises around the audio output device. The audio output device may include a controller configured to analyze the detected sound and noises and may adjust audio settings of the audio output device accordingly. In an embodiment, the controller may analyze and recognize that the detected noise is another person's attempt to get the user's attention or to talk to the user, such as audio sound of the user's name. In response, the controller may stop the audio output or lower the volume of the audio output.

Abstract: This application relates to an audio processing method, an electronic device, and a storage medium. The method includes: displaying a target audio clip and corresponding target text information having a mapping relationship between a location of an audio segment in the target audio clip and a location of text information in the corresponding target text information; receiving, a selection of a location in the corresponding target text information as a to-be-processed text location; matching a to-be-processed audio location of an audio segment that has the mapping relationship with the to-be-processed text location; and processing the target audio at the to-be-processed audio location to generate an updated target audio clip, and updating the corresponding target text information at the to-be-processed text location to generate updated target text information; and displaying the updated target audio clip and the updated target text information.

Abstract: A method and system for operating a multitenancy telephony system including receiving a plurality of call requests associated with an application server; selecting a subset of the received call requests using a load balancer; assigning each selected call request to one of a plurality of resources creating, on the assigned resource, a call session for each selected call request; and coordinating the created call sessions resulting from the selected subset of the received call requests associated with the application server.

Abstract: A guide robot can include a travel part to move the guide robot, a touch screen and a camera, a sensor to detect an approach of a user, and a voice reception part to receive a voice. The guide robot further includes a controller to display at least one digital signage while the guide robot is traveling, in response to detecting the approach of the user, stop the traveling of the guide robot and transition the camera from a deactivated state to an activated state, and detect a face and a face angle of the user. Also, in response to determining that the user intends to use the guide robot, the controller can trigger a voice conversation mode by activating the voice reception part, stopping the display of the at least one digital signage and outputting usage guide information for the voice conversation mode.

Abstract: Disclosed is a synchronous and dynamic loading method in electro-magneto-thermo-mechanical multi-field coupling conditions. The method comprises the following steps: applying maximum pulse current to a test object by a pulse power supply to realize loading in extreme electric field and magnetic field conditions; meanwhile, generating a large amount of friction heat by the high-speed rotation of a rotating body and the friction of the test object to realize loading in an extreme-temperature field combined with a large amount of Joule heat and arc heat; synchronously applying pressure to the rotating body by a pressure device to realize loading of extreme force combined with the gravity of the rotating body and the friction force between the rotating body and the test object.

Abstract: Techniques are disclosed herein for focused training of language models and end-to-end hypertuning of the framework. In one aspect, a method is provided that includes obtaining a machine learning model pre-trained for language modeling, and post-training the machine learning model for various tasks to generate a focused machine learning model. The post-training includes: (i) training the machine learning model on an unlabeled set of training data pertaining to a task that the machine learning model was pre-trained for as part of the language modeling, and the unlabeled set of training data is obtained with respect to a target domain, a target task, or a target language, and (ii) training the machine learning model on a labeled set of training data that pertains to another task that is an auxiliary task related to a downstream task to be performed using the machine learning model or output from the machine learning model.

Abstract: A method includes accessing, using at least one processor of an electronic device, a machine learning model. The machine learning model is a trained student model that is trained using audio samples in a plurality of accent types. The method also includes receiving, using the at least one processor, an audio input from an audio input device. The method further includes providing, using the at least one processor, the audio input to the trained student model. The method also includes receiving, using the at least one processor, an output from the trained student model including frame-level probabilities associated with the audio input. In addition, the method includes instructing, using the at least one processor, at least one action based on the frame-level probabilities associated with the audio input.

Abstract: An electronic control device of an avionics system for implementation of a critical avionics function, comprising: a module for receiving a voice instruction signal; a speech recognition module configured to transform the voice signal into a textual transcript; a processing module configured to associate the textual transcript with at least one action to be performed; a monitoring system comprising: a control module configured to check whether the textual transcript and/or the action to be performed is consistent if and only if: a) the textual transcript and/or the action to be performed is consistent with the expected syntax, b) the textual transcript and/or the action to be performed is consistent with the expected lexical field, and c) the textual transcript and/or the action to be performed is consistent with the current context, a module for generating an associated command only if no inconsistencies are detected.

Abstract: Techniques are disclosed for coordinating interactions between a user device and a plurality of accessory devices. In one example, a user device receives information identifying one or more accessory devices in communication with the user device. The user device may implement accessory interaction instances for each of the identified accessories. A first accessory interaction instance can be associated with a first accessory among the identified accessories and receive a first audio input from the first accessory corresponding to a user request. The first accessory interaction instance can process a portion of the received audio input and receive a first response from a server computer. The user device may then transmit the first response to the first accessory device.

Abstract: A system and method for determining user temperament. The system includes at least one processor, a communication interface communicatively coupled to the at least one processor, and a memory device storing executable code. The executable code causes the processor(s) to receive an incoming call from a user, receive audible language from the user, determine from the language properties such as: volume, content, inflection, and/or pitch, input the determined properties into an algorithm configured to rank temperament, and receive as output from the algorithm a ranking of the user's temperament. Additionally, the executable code may cause the processor to transfer the incoming call to a representative and relay the user's temperament ranking to the representative. The representative chosen for the user may be selected based on the user's temperament ranking. The system may furnish the representative with a script that correlates with the user's temperament ranking.

Abstract: Methods, systems, and computer storage media for providing predicted task-actions for tasks from a plurality of task applications associated with a task management system. In operation, a task associated with a task application is accessed at a task-action prediction engine. An intent of the task is determined. The intent of the task is determined based on task features and an intent prediction machine learning model that is trained to predict intents associated with natural language text. Based on the intent of the task, a predicted task-action is selected based on a task-action computation model that is associated with predefined task-actions and predefined intent-task-categories. The predicted task-action can be communicated to a task management client that causes display of the predicted task-action in combination with a predicted task-action interface element associated with executing the predicted task-action.

Abstract: Aspects of the disclosure relate to preserving text and speech semantics using real-time virtual corpora in live interactive chatbot sessions. A computing platform may receive user input and may determine whether characteristics of the input are identifiable. Based on determining the characteristics are identifiable, the computing platform may populate the characteristics using a global corpora and may use the populated characteristics to generate a response to the input. Alternatively, based on determining the characteristics are unidentifiable, the computing platform may determine whether the same or a similar input is stored in a virtual corpora. If the same or a similar input is stored in the virtual corpora, the computing platform may generate a response to the input using the virtual corpora. Alternatively, if the same or a similar input was not previously received, the computing platform may gather information from the user to generate a response to the input.

Abstract: A system and method for customizing communication processing within a communication platform that includes configuring a service handler, which comprises of at least setting execution code of the service handler; setting an association between the service handler and a communication event of the communication platform; detecting the occurrence of the communication event of a communication facilitated through the communication platform; invoking the service handler in response to the event, which comprises: executing the execution code in a serverless execution environment and generating a service handler response; and augmenting the communication in accordance with the service handler response.