Abstract: A method for protecting a biometric template, comprising the steps of: retrieving an original vector (V) representing said biometric template, said vector comprising a plurality of original elements (v1, v2, . . . vi, . . . , vn); mapping at least some elements from said original vector to a protected vector (P) comprising a plurality of protected elements (p1, p2, . . . pi, . . . , pn?m+1), the mapping being based on multivariate polynomials defined by m user-specific coefficients (C) and exponents (E).

Abstract: The present invention discloses a speaker device. The speaker device comprises a speaker box, a heat source, and a heat conducting member; the speaker box comprises a housing, a speaker, a sound-guiding channel, and a heat conducting plate; the speaker includes a diaphragm, the first receiving space is divided into a front sounding chamber and a back chamber, and the sound-guiding channel communicates the front sounding chamber with the outside and forms a front chamber together with the front sounding chamber; the housing having an opening penetrating therethrough, the opening communicates with the front chamber, and the heat conducting plate covers on the opening, an adjusting hole penetrates through the housing and communicates with the back chamber; an valve covers on the adjusting hole and a controller for outputting a control signal to control the valve to open or to close.

Abstract: Techniques are described herein for training and evaluating machine learning (ML) models for document processing computing applications using generalized vocabulary tokens. In some embodiments, an ML system determines a set of tokens for non-textual content in a plurality of documents. The ML system generates a fixed-length vocabulary that includes the set of tokens for the non-textual content. The ML system further generates for each respective document in a training dataset of documents, a respective feature vector based at least in part on which tokens in the fixed-length vocabulary occur in the respective document. The ML system trains a ML model based at least in part on the respective feature vector for each respective document in the training dataset.

Abstract: In response to determining that a native language of a first user is different from a target language of a message to be transmitted by the first user to a second user, a translation model based on a plurality of language efficacies of the first user is created. An optimal action associated with a translation of the message from the native language to the target language is determined based on the created model and a language efficacy of the first user in the native language. The determined optimal action is performed. The message translation comprising the performed optimal action is transmitted to the second user.

Abstract: The present disclosure provides an artificial intelligence apparatus which inputs first language data into a machine translation model to economically train a natural language understanding model of a second language and obtains second language data corresponding to the first language data to train the natural language understanding model.

Abstract: Techniques and structures to facilitate conversion of a workflow process is disclosed. The techniques include receiving an image, identifying one or more objects included in the image, identifying one or more properties associated with each of the one or more objects, generating a matrix including data including the identified objects and associated properties and processing the matrix at a machine learning model to determine whether the image is to be translated based on a determination that one or more objects and associated properties within the image are required to be translated.

Abstract: Systems and methods are configured for predictive prompt generation for an interaction between a party and an automated prompt system. In various embodiments, metadata is received on the interaction and provided as input to a multi-label predictive model to generate interaction probabilities for a plurality of prompt information data objects. Each probability generated by the predictive model provides a likelihood a particular information data object needs to be provided to the first party during the interaction. Accordingly, one or more of the prompt information data objects are identified based on the probability of each piece found in the one or more prompt information data objects that satisfy a set threshold and one or more notifications are provided so that the automated prompt system provides at least one of the prompt information data objects to the first party during the interaction.

Abstract: In one embodiment, a method includes receiving a query from a user from a client system associated with the user, accessing a knowledge graph comprising a plurality of nodes and edges connecting the nodes, wherein each node corresponds to an entity and each edge corresponds to a relationship between the entities corresponding to the connected nodes, determining one or more initial entities associated with the query based on the query, selecting one or more candidate nodes by a conversational reasoning model from the knowledge graph corresponding to one or more candidate entities, respectively, wherein each candidate node is selected based on the nodes corresponding to the initial entities, dialog states associated with the query, and a context associated with the query, generating a response based on the initial entities and the candidate entities, and sending instructions for presenting the response to the client system in response to the query.

Abstract: Method starts with a processor receiving configuration settings including an identified task, a relationship data, and a criticality value. Processor initializes a communication session with an agent client device. The communication session is between a virtual caller associated with the system and the agent client device. Processor then processes an audio signal of the communication session to generate an agent utterance and generates a transcribed agent utterance based on the agent utterance using a speech-to-text processor. Processor generates a virtual caller utterance using a task-specific virtual caller neural network associated with the identified task. The virtual caller utterance can be generated based on the transcribed agent utterance. Processor then causes the virtual caller utterance to be played back in the communication session to the agent client device. Other embodiments are disclosed herein.

Abstract: Techniques are described for managing out of service telephone numbers (TNs) before returning them to active service. A telephone number profile score may be calculated for an out of service TN entering an aging protocol, the aging protocol comprising a pre-determined number of days the TN must remain out of service. The out of service TN may then be assigned a profile based on its telephone number profile score, the profile indicative of how quickly the TN will achieve one or more quality thresholds. The number of days from entry into the aging protocol it will take the TN to achieve the one or more quality thresholds based on its assigned profile may be determined. The quality threshold status of the TN may then be updated each time it crosses one of the quality thresholds. Once it completes the aging process, the TN may be placed into an inventory of TNs eligible for active service.

Abstract: A mechanism is provided to implement an abnormal entity detection mechanism that facilitates detecting abnormal entities in real-time response systems through weak supervision. For each first intent from an entity labeled workspace that matches a second intent in labeled chat logs, when the entity score associated with each first entity or second entity is above a predefined significance level the first entity or the second entity is recorded. For each first intent from the entity labeled workspace that matches the second intent in the labeled chat logs: responsive to the first entity being recorded and the second entity failing to be recorded, that first entity is removed from the training data as being mistakenly included; or, responsive to the second entity being recorded and the first entity failing to be recorded, that second entity is added as a potential business case to the training data.

Abstract: Methods, systems, and software are disclosed for connecting insurance company customers contacting the insurance company to independent insurance agents.

Abstract: Methods, devices, and systems for contactless cough detection and attribution are presented herein. Audio data may be received using a microphone. A cough may be identified as having occurred based on the received audio data. Radar data may be received indicative of reflected radio waves from a radar sensor. A state analysis process may be performed using the received radar data. The detected cough may be attributed to a particular user based at least in part on the state analysis process performed using the radar data.

Abstract: Embodiments for managing chatbots are provided. A set of documents is received. A plurality of entities are identified within the set of documents. At least one of the plurality of entities is selected based on a rareness criteria. Contextual data associated with each of the selected at least one of the plurality of entities is identified within the set of documents. At least one question-answer (QA) pair associated with each of the selected at least one of the plurality of entities is generated based on the identified contextual data.

Abstract: When a chatbot application is in a mode of live chat conversation with a user, a trained intent classifier determines an intent that underlies a first live question received by the trained intent classifier from the user. A trained next predictor receives the intent from the intent classifier. The trained next intent predictor generates a set of predicted next intents responsive to receiving the intent. A trained re-ranker selects at least one of the predicted next intents responsive to receiving the set of predicted next intents. A question selection engine sends at least one suggested question to the user responsive to receiving the at least one predicted next intent. As a result of the above, the chatbot application provides to the user at least one suggested next question the user may wish to ask in response to the first question from the user.

Abstract: The present disclosure provides methods and systems that may be used for providing quality control for audio samples. The audio samples may be speech samples of a user. The user may be participating in an audio interview.

Abstract: A conference participant is alerted as to an event during a conference responsive to a determination that a focus of the conference participant is other than on the conference. During the conference, an event associated with the conference participant is detected based on a real-time transcription of the conference. For example, the event may relate to a topic relevant to the conference participant or a request associated with a name of the conference participant. A determination is made that a focus of the conference participant is other than on the conference based on information associated with a device of the conference participant, such as input received from a camera associated with the device or a setting of an audio output device associated with the device. Based on that determination and the detected event, output is presented to alert the conference participant as to the event.

Abstract: Technology is disclosed for proactively enriching content of electronic communications through predicting user intents. User-composed content within an electronic communication is observed and used to predict one or more intents each associated with a task related to the communication and to be completed in creating the communication. Task-queries may be generated based on the detected intents and entities extracted from the user-composed content, contextual information, or other information related to the user, such as a user-knowledge database. User selection of a suggested task-query automatically executes the task-query, and the results are provided to the user. Each result may include an action involving adding content to or supplementing the content of the electronic communication. A conversational digital assistant may further be invoked to interact with the user to confirm and/or modify suggested task-queries and results of selected task-queries.

Abstract: A speech interface device is configured to update personalized data in local memory for responding to user speech. The speech interface device may receive compiled personalized data from a remote system, or raw personalized data from local devices in the environment of the speech interface device, the raw personalized data being compiled locally on the speech interface device. The compiled personalized data is received by an artifact manager of the speech interface device for storage in the memory of the speech interface device. A local speech processing component of the speech interface device may use the compiled personalized data when processing user speech in order to understand a spoken form of a word or phrase that is associated with a user. This allows for responding to personal user speech, even when the speech interface device is unable to communicate with a remote speech processing system over a wide area network.

Abstract: Techniques are provided for spatial audio streaming using adaptive bitrate encoding and decoding in bandwidth limited virtual reality applications. A methodology implementing the techniques according to an embodiment includes calculating weight factors for audio channels associated with each of a plurality of microphones. The weight factors are based on the position and head angle, relative to the microphones, of a user wearing a head mounted display (HMD). The method also includes encoding and decoding the audio channels, for transmission to the HMD, at a selected bitrate, the bitrate based on the weight factor for the audio channel. The method further includes applying an adaptive gain compensation to each of the received and decoded audio channels, the gain compensation also based on the weight factor. The method further includes mixing the resulting gain compensated channels to generate a stereo audio signal to be played through speakers of the HMD.