Abstract: A method for training an artificial neural network-based speech recognition model is disclosed. In the method for training an artificial neural network-based speech recognition model, a user's speech is learned by using target data representing features and non-target data representing non-features as random inputs and outputs, and then the user's speech is recognized under a noise situation. A method for training an artificial neural network-based speech recognition model and speech recognition device of the present disclosure can be associated with artificial intelligence modules, drones (unmanned aerial vehicles (UAVs)), robots, augmented reality (AR) devices, virtual reality (VR) devices, devices related to 5G service, etc.

Abstract: The various implementations described herein include methods and systems for determining device leadership among voice interface devices. In one aspect, a method is performed at a first electronic device of a plurality of electronic devices, each having microphones, a speaker, processors, and memory storing programs for execution by the processors. The first device detects a voice input. It determines a device state and a relevance of the voice input. It identifies a subset of electronic devices from the plurality to which the voice input is relevant. In accordance with a determination that the subset includes the first device, the first device determines a first score of a criterion associated with the voice input and receives second scores of the criterion from other devices in the subset. In accordance with a determination that the first score is higher than the second scores, the first device responds to the detected input.

Abstract: Systems and methods are described herein for providing media guidance. Control circuitry may receive a first voice input and access a database of topics to identify a first topic associated with the first voice input. A user interface may generate a first response to the first voice input, and subsequent to generating the first response, the control circuitry may receive a second voice input. The control circuitry may determine a match between the second voice input and an interruption input such as a period of silence or a keyword or a phrase, such as “Ahh,”, “Umm,”, or “Hmm.” The user interface may generate a second response that is associated with a second topic related to the first topic. By interrupting the conversation and changing the subject from time to time, media guidance systems can appear to be more intelligent and human.

Abstract: Voice command matching during testing of voice-assisted application prototypes for languages with non-phonetic alphabets is described. A visual page of an application prototype is displayed during a testing phase of the application prototype. A speech-to-text service converts a non-phonetic voice command spoken in a language with a non-phonetic alphabet, captured by at least one microphone during the testing phase of the application prototype, into a non-phonetic text string in the non-phonetic alphabet of the voice command. A phonetic language translator translates the non-phonetic text string of the voice command into a phonetic text string in a phonetic alphabet of the voice command. A comparison module compares the phonetic text string of the voice command to phonetic text strings in the phonetic alphabet of stored voice commands associated with the application prototype to identify a matching voice command. A performance module performs an action associated with the matching voice command.

Abstract: A method of generating text having related purposes using a generative adversarial network (GAN) includes receiving a limited dataset including real data with related cognitive value types (types). The method includes applying loss functions to portions of the real data. The portions of the real data are each identified as having one of the types. The loss functions ensure alignment of the portions with corresponding types. The method includes embedding the real data into an encoder output that includes an embedded vector for the cognitive value types. The method includes generating an additional parameter set supplemental to the real data and configured to enhance an expressiveness of a model. The method includes generating statements based on the additional parameter set and the encoder output. The statements include a style of one of the cognitive value types and are related to a common issue addressed by the GAN.

Abstract: Systems and processes for unified language modeling are provided. In accordance with one example, a method includes, at an electronic device with one or more processors and memory, receiving a character of a sequence of characters and determining a current character context based on the received character of the sequence of characters and a previous character context. The method further includes determining a current word representation based on the current character context and determining a current word context based on the current word representation and a previous word context. The method further includes determining a next word representation based on the current word context and providing the next word representation.

Abstract: A system includes one or more memory devices storing instructions, and one or more processors configured to execute the instructions to perform steps of providing automated natural dialogue with a customer. The system may generate one or more events and commands temporarily stored in queues to be processed by one or more of a dialogue management device, an API server, and an NLP device. The dialogue management device may create adaptive responses to customer communications using a customer context, a rules-based platform, and a trained machine learning model.

Abstract: Systems and methods for generating best next communication policies, for a time step of an exchange of electronic documents, fit over historical exchanges, optimizing to maximize a probability of achieving a quantified objective leveraging weighted sampling. In a preferred embodiment an electronic document is segmented whereby each constituent segment is deconstructed as a composition of custom expression varieties, pre-defined to enable fulfilment of an objective within a theme of correspondence, associating each expression with a semantic vector. A set of expression extraction models is trained independently and then a second set with knowledge of parallel label predictions, iterating to convergence. The expression compositions and associated semantic vectors are combined into a single vector for each segment. The segment vectors are appended onto profile vectors for the exchange parties, yielding a time series of profile-content vectors.

Abstract: Embodiments of the present disclosure provide voice interaction method and device. The method includes: determining whether a first query statement currently received is a query statement first received within a preset time period; if not, obtaining a second query statement, where the second query statement is a query statement lastly received before receiving the first query statement; obtaining a third sentence vector according to a first sentence vector of the first query statement and a second sentence vector of the second query statement; and obtaining, from a bottom corpus, a first question and answer result corresponding to a fourth sentence vector a similarity between which and the third sentence vector satisfies a preset condition, and returning the first question and answer result. The method provided in the embodiment can return a bottom reply irrelevant to the query statement to the user, thereby improving the user experience.

Abstract: Techniques for routing items addressed to an unstructured address are described. One embodiment includes receiving an order for delivery of a first package, the order specifying a first address that does not comply with a defined address format. The first address is processed using one or more hybrid machine learning algorithms to determine a Normalized Delivery Location (NDL) associated with the first address. A sorting zone that encompasses the NDL is determined. The sorting zones correspond to a predefined geographic region. Embodiments facilitate transport of the first package to a physical shipping location within the predefined geographic region.

Abstract: According to an embodiment, a conference information accumulating apparatus is for accumulating conference information. The apparatus includes a generator and a calculator. The generator is configured to generate a user interface screen either for creating minutes of a conference based on the conference information or for viewing the created minutes. The calculator is configured to calculate a correlation between a written text that is a unit in which the minutes are written and the conference information, based on a predetermined operation performed using the user interface screen by a minutes creator. The generator generates, upon detection of the conference information that is correlated with the written text, the user interface screen enabling a reference to the conference information.

Abstract: An audio signal encoding method is provided that comprises collecting audio signal samples, determining sinusoidal components in subsequent frames, estimating amplitudes and frequencies of the components for each frame, merging the obtained pairs into sinusoidal trajectories, splitting particular trajectories into segments, transforming particular trajectories to the frequency domain by way of a digital transform performed on segments longer than the frame duration, quantization and selection of transform coefficients in the segments, entropy encoding, outputting the quantized coefficients as output data, wherein segments of different trajectories starting within a particular time are grouped into Groups of Segments, and the partitioning of trajectories into segments is synchronized with the endpoints of a Group of Segments.

Abstract: A hub device, a multi-device system including the hub device, and a method of operating the same may include: converting, by the hub device, received voice input into text; identifying, by the hub device, a device capable of performing an operation corresponding to the text; identifying which device stores a function determination model corresponding to the device capable of performing the operation corresponding to the text, from among the hub device, and a plurality of other devices connected to the hub device; and based on the identified device that stores the function determination model being a device that is different from the hub device, transmitting at least part of the text to the identified device.

Abstract: The present invention relates to a washing machine and a server system that recommend a laundry course and washing tip information in consideration of information on a kind of laundry and a degree of contamination inputted by using artificial intelligence through speech recognition, and a method for controlling such washing machine and server system. The present invention extracts a cloth-word indicating a kind of clothes of laundry and a stain-word indicating a kind of contaminant. Then, the present invention determines a laundry course in consideration of the cloth-word. Then, washing tip information on the stain-word is searched in a pre-stored database and the searched washing tip information is provided to a user. As a result, the user can be notified of a washing method capable of effectively removing the contaminant.

Abstract: Systems, methods, and devices for speech transformation and generating synthetic speech using deep generative models are disclosed. A method of the disclosure includes receiving input audio data comprising a plurality of iterations of a speech utterance from a plurality of speakers. The method includes generating an input spectrogram based on the input audio data and transmitting the input spectrogram to a neural network configured to generate an output spectrogram. The method includes receiving the output spectrogram from the neural network and, based on the output spectrogram, generating synthetic audio data comprising the speech utterance.

Abstract: A non-transitory computer-readable storage medium storing a program that causes a computer to execute a process, the process including generating index information regarding a first string data including a plurality of words or phrases, the index information indicating appearance positions of each of characters, setting flag information to the index information, the flag information indicating head positions plurality of words or phrases and end positions of each of the plurality of words or phrases, and extracting, from a second string data, a plurality of dividable words or phrases corresponding to the plurality of words or phrases by using the index information to which the flag information is set.

Abstract: One embodiment provides a method, including: receiving, at an information handling device, an indication to provide output; identifying that the output comprises confidential information; determining, using a processor, that the output may be heard by at least one other user; and altering, responsive to the determining, the output to conceal the confidential information. Other aspects are described and claimed.

Abstract: A technique for semantic search and retrieval that is event-based, wherein is event is composed of a sequence of observations that are user speech or physical actions. Using a first set of conversations, a machine learning model is trained against groupings of utterances therein to generate a speech act classifier. Observation sequences therein are organized into groupings of events and configured for subsequent event recognition. A set of second (unannotated) conversations are then received. The set of second conversations is evaluated using the speech act classifier and information retrieved from the event recognition to generate event-level metadata that comprises, for each utterance or physical action within an event, one or more associated tags. In response to a query, a search is performed against the metadata. Because the metadata is derived from event recognition, the search is performed against events learned from the set of first conversations.

Abstract: Text can be presented with speech indicators generated by a cognitive system by processing the text. The speech indicators can indicate recommended speech characteristics to be exhibited by a user while the user generates spoken utterances representing the text. Data indicating at least one user input changing at least one of the speech indicators from a first state as originally presented to a second state can be received. In response, a value indicating a level of change made to the at least one of the speech indicators can be determined. At least one parameter used by the cognitive system to select the speech indicators can be modified or created based on the value indicating the level of change made to the at least one of the speech indicators.

Abstract: A method for detecting a hotword includes receiving a sequence of input frames that characterize streaming audio captured by a user device and generating a probability score indicating a presence of a hotword in the streaming audio using a memorized neural network. The network includes sequentially-stacked single value decomposition filter (SVDF) layers and each SVDF layer includes at least one neuron. Each neuron includes a respective memory component, a first stage configured to perform filtering on audio features of each input frame individually and output to the memory component, and a second stage configured to perform filtering on all the filtered audio features residing in the respective memory component. The method also includes determining whether the probability score satisfies a hotword detection threshold and initiating a wake-up process on the user device for processing additional terms.