Abstract: Embodiments are directed to summarizing conversational speech. Conversation segments may be provided based on a conversation stream and segmentation models. Summarization models may be determined based on characteristics of the conversation segments. Summarization information may be generated for each of the conversation segments based on the summarization models such that the summarization information includes a text-based summarization of the conversation segment. Summarization profiles may be generated for the conversation segments based on the summarization information such that each summarization profile is associated with quality scores. Summarization models may be modified based on the summarization profiles and the associated quality scores such that the summarization profiles are updated based on the modified summarization models. Modified summarization models and the updated summarization profiles may be employed to provide reports to a user.

Abstract: The present disclosure discloses a control method and a control apparatus for speech interaction. The detailed implementation solution of the control method for the speech interaction includes: collecting an audio signal; detecting a wake-up word in the audio signal to obtain a wake-up word result; and playing a prompt tone and/or executing a speech instruction in the audio signal based on the wake-up word result.

Abstract: Disclosed herein is an NLP system that is able to extract meaning from a natural language message using improved parsing techniques. Such an NLP system can be used in concert with an NLG system to interactively interpret messages and generate response messages in an interactive conversational stream. The parsing can include (1) named entity recognition that contextualizes the meanings of words in a message with reference to a knowledge base of named entities understood by the NLP and NLG systems, (2) syntactically parsing the message to determine a grammatical hierarchy for the named entities within the message, (3) reduction of recognized named entities into aggregations of named entities using the determined grammatical hierarchy and reduction rules to further clarify the message's meaning, and (4) mapping the reduced aggregation of named entities to an intent or meaning, wherein this intent/meaning can be used as control instructions for an NLG process.

Abstract: This disclosure describes a solution to identify that a meaningful event has occurred in a person's life. Once identified, data and content related to the event can be collected and stored in a database. This data and content can be used to offer future virtual reality (VR) experiences and content. A person can be equipped with a smart device such as a smartphone, smart watch, or other wearable that can further be equipped with a tracking app that collects data from these devices and from other sources, for instance, over the internet. The tracking app can continuously collect such data and use various algorithms to make a determination as to whether the person is experiencing a meaningful life event.

Abstract: Systems, apparatuses, and methods are described for assisting speech recognition processing. If speech recognition processing of speech input by an individual does not yield a recognized result, for example, if speech is from an individual with compromised speech, an indication may be sent to a device associated with another person that can assist. The other person may provide, via that device, additional input that indicates the meaning of the speech input. Based on this additional input, an assisted speech recognition result may be determined.

Abstract: It is an object to successfully absorb a difference in characteristics to be taken into consideration between languages and implement common named entity extraction in a processing system.

Abstract: Systems and methods are disclosed herein for training a control system based on prior audio inputs. The disclosed systems and methods receive a non-lexical or interjectional audio input. State change indications are also received and stored by the system within a predefined period of time starting from the time the system received the audio input. The system then receives a subsequent audio input. If the audio inputs of both the audio input and the subsequent audio input match, and contextual information for the audio input and the subsequent audio input match, the system stores a match association, comprising a confidence factor, for the subsequent audio input to the audio input in the associative data structure. If the confidence factor is greater than a preconfigured confidence level, the system executes one or more functions based on stored state change indications.

Abstract: The present disclosure describes a conversation facilitation system for facilitating conversation-based social interactions to improve senior health, one or more operations and functions being efficiently achieved via this system comprise: receiving a dialog act of a conversation, applying natural language understanding (NLU) processing on the dialog act, computing a conversation metric, and generating a result of the conversation to conclude the conversation based on the conversation metric.

Abstract: Aspects of the present disclosure involve a system comprising a computer-readable storage medium storing a program and method for displaying augmented reality content. The program and method provide for receiving, by a device, speech input to select augmented reality content for display, determining at least one keyword included in the speech input; identifying, from plural augmented reality content items, an augmented reality content item corresponding to the at least one keyword; and displaying the augmented reality content item with an image captured by a camera of the device.

Abstract: In an approach to providing contextual justification for a virtual assistant response, one or more computer processors receive a first voice command from a user. One or more computer processors determine one or more boundary conditions associated with the first voice command. Based on the one or more boundary conditions, one or more computer processors determine a first response to the first voice command and a contextual justification of the first response. One or more computer processors respond to the user with the response to the first voice command and the contextual justification of the response.

Abstract: An equipment and a method for configuring a service on an equipment. A method includes receiving a first voice input from a user to configure an equipment with a service. The equipment is configured with a voice-bot to interact with the user. The method also includes validating the first voice input, initiating configuration of the service and outputting a first voice response based on the validation of the first voice input. The method includes receiving a second voice input from the user in response to the first voice response and validating the second voice input. The method includes outputting a second voice response based on the validation of the second voice input and configuring the service on the equipment based on the voice inputs from the user.

Abstract: System, electronic device, and related methods, in particular a method of operating a system comprising an electronic device is disclosed, the method comprising obtaining one or more audio signals including a first audio signal of a first conversation; determining first speaker metric data of a first speaker based on the first audio signal, the first speaker metric data including first primary speaker metric data; detecting a termination of the first conversation; in accordance with detecting the termination of the first conversation, determining a first post-conversation representation based on the first speaker metric data; and outputting, via the interface of the electronic device, the first post-conversation representation.

Abstract: An image forming apparatus includes a plurality of devices configured to perform different job processing, a control unit configured to control job processing performed by each device, a reception unit configured to receive an instruction for causing the control unit to shift to a state where the job processing is capable of being performed, and a power control unit configured to, when the control unit is shifted to a stand-by state in response to receiving the instruction, supply a power to a device specified based on a job processing function corresponding to an initial screen to be displayed.

Abstract: Devices and techniques are generally described for pre-caching of speech processing feature data. In various examples, first data indicating source data is received from a first speech processing component. The source data may be used to generate first feature data. In various examples, a first request to process first input data is received. A second speech processing component may generate the source data during processing of the first input data. The first feature data may be generated using the source data. The first feature data may be sent to the first speech processing component. In some examples, the first speech processing component may store the first feature data in a first cache local to the first speech processing component.

Abstract: A method and system for acoustic model conditioning on non-phoneme information features for optimized automatic speech recognition is provided. The method includes using an encoder model to encode sound embedding from a known key phrase of speech and conditioning an acoustic model with the sound embedding to optimize its performance in inferring the probabilities of phonemes in the speech. The sound embedding can comprise non-phoneme information related to the key phrase and the following utterance. Further, the encoder model and the acoustic model can be neural networks that are jointly trained with audio data.

Abstract: Speaker diarization techniques that enable processing of audio data to generate one or more refined versions of the audio data, where each of the refined versions of the audio data isolates one or more utterances of a single respective human speaker. Various implementations generate a refined version of audio data that isolates utterance(s) of a single human speaker by generating a speaker embedding for the single human speaker, and processing the audio data using a trained generative model—and using the speaker embedding in determining activations for hidden layers of the trained generative model during the processing. Output is generated over the trained generative model based on the processing, and the output is the refined version of the audio data.

Abstract: Network microphone devices configured to detect keywords can include microphones for capturing sound samples. Features can be extracted from the sound samples by storing the sound samples in a first portion of a dynamic-access memory block, performing first computations based on spectral coefficients of the sound samples using a second portion of the memory block, and storing results of the first computations as extracted features in a third portion of the memory block. The second and third portions of the memory block can be designated as temporary memory. The extracted features are then processed using a neural network by storing the extracted features in a fourth portion of the memory block, performing second computations on the extracted features using the temporary memory, the second computations comprising computing at least one layer of the neural network, and storing an output of the neural network as a classification in the temporary memory.

Abstract: Implementations set forth herein relate to initializing performance of an automated assistant routine and/or dismissing an alarm pre-emptively according to satisfaction of one or more conditions. A condition can be satisfied by a user acknowledging the alarm when the alarm is going off, or causing the alarm to be dismissed prior to a time at which the alarm was scheduled for. The user can cause the alarm to be dismissed pre-emptively by interacting with the automated assistant prior to the time the alarm was scheduled for and/or interacting with a device, which is known to the automated assistant, prior to the time that the alarm was scheduled for. In this way, actions that cause an alarm to be dismissed can be recognized and used to initialize other processes, such as an automated assistant routine, thereby reducing a number of inputs needed from a user.

Abstract: Systems and methods for real-time correction of an accent in a speech audio signal are provided. A method includes dividing the speech audio signal into a stream of input chunks, an input chunk from the stream of input chunks including a pre-defined number of frames of the speech audio signal, extracting, by an acoustic features extraction module from the input chunk and a context associated with the input chunk, acoustic features, the context is a pre-determined number of the frames preceding the input chunk in the stream; extracting, by a linguistic features extraction module from the input chunk and the context, linguistic features, receiving a speaker embedding for a human speaker, providing the speaker embedding, the acoustic features, and the linguistic features to a synthesis module to generate a melspectrogram with a reduced accent, providing the melspectrogram to a vocoder to generate an output chunk of an output audio signal.

Abstract: An ASR model includes a first encoder configured to receive a sequence of acoustic frames and generate a first higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames. The ASR model also includes a second encoder configured to receive the first higher order feature representation generated by the first encoder at each of the plurality of output steps and generate a second higher order feature representation for a corresponding first higher order feature frame. The ASR model also includes a decoder configured to receive the second higher order feature representation generated by the second encoder at each of the plurality of output steps and generate a first probability distribution over possible speech recognition hypothesis. The ASR model also includes a language model configured to receive the first probability distribution over possible speech hypothesis and generate a rescored probability distribution.