Abstract: Systems and methods for using speech recognition technologies to facilitate retail store checkout are disclosed. According to certain aspects, an electronic device may detect a user's speech and analyze the speech to identify a set of matching items that may correspond to items being purchased by a customer. The electronic device may display, via a user interface, the set of matching items as well as a code or identification associated with the set of matching items. The user may interface with a point of sale system to input a code for a desired item, and the point of sale system may add the desired item to an order and may facilitate a checkout for the order.

Abstract: Described herein are systems and methods for compressing and speeding up dense matrix multiplications as found, for examples, in the fully connected and recurrent layers of neural networks for embedded large vocabulary continuous speech recognition (LVCSR). For compression, trace norm regularization technique embodiments were introduced and studied for training low rank factored versions of matrix multiplications. Compared to standard low rank training, the methods more consistently lead to good accuracy versus number of parameter trade-offs and can be used to speed-up training of large models. Faster inference may be further enabled on ARM processors through kernels optimized for small batch sizes, resulting in speed ups over the currently used library. Beyond LVCSR, the techniques are also generally applicable to embedded neural networks with large fully connected or recurrent layers.

Abstract: A centralized and robust threat assessment tool is disclosed to perform comprehensive analysis of previously-stored and subsequent communication data, activity data, and other relevant information relating to inmates within a controlled environment facility. As part of the analysis, the system detects certain keywords and key interactions with the dataset in order to identify particular criminal proclivities of the inmate. Based on the identified proclivities, the system assigns threat scores to inmate that represents a relative likelihood that the inmate will carry out or be drawn to certain threats and/or criminal activities. This analysis provides a predictive tool for assessing an inmate's ability to rehabilitate. Based on the analysis, remedial measures can be taken in order to correct an inmate's trajectory within the controlled environment and increase the likelihood of successful rehabilitation, as well as to prevent potential criminal acts.

Abstract: Disclosed is a method of phonetically encoding a text document. The method comprises providing, for a current word in the text document, a phonetically equivalent encoded word comprising one or more syllables, each syllable comprising a sequence of phonemes from a predetermined phoneme set, the sequence being phonetically equivalent to the corresponding syllable in the current word, and adding the phonetically equivalent encoded word or the current word at a current position in the phonetically encoded document, Each phoneme in the phoneme set is associated with a base grapheme that is pronounced as the phoneme in one or more English words.

Abstract: This disclosure describes converting computer-executable predicate-argument structures for a specific field to field-specific predicated-argument structures to improve execution. In some implementations, a method can be performed by one or more processors of a computing device, and can include receiving one or more predicate-argument structures (PASs) associated with taxation-specific text and converting the one or more PASs into one or more tax-specific predicate-argument structures (TPASs). Converting the one or more PASs to one or more TPASs may include one or more of: defining terms in a segment based on a definition of the term from a different segment or line description (including from a different document); reordering nodes, replacing nodes, or removing nodes of a segment (such as based on one or more single segment tree traversal rules); or combining multiple PASs for multiple segments of a single line description based on one or more multiple segment tree traversal rules.

Abstract: A system for providing intelligent part of speech processing of complex natural language is disclosed. The system identifies a multiword concept from an input and replaces the multiword concept with a token to be tagged as a desired part of speech. The system passes the modified text including the token to a part-of-speech tagger to tag each word in the text with the appropriate part-of-speech. The system may replace the token with the original text that the token was utilized to replace so that the original intent of the text is evident. The system may analyze the tagged text to generate analyses and interpretations associated with the input. When multiple multiword concepts are identified, the system may evaluate them by computing scores for each of the multiword concepts that may be replaced with tokens, for each of the modified texts including the tokens, or for any interpretations and analyses thereof.

Abstract: Described herein are systems, methods, and apparatus for determining audio context between an audio source and an audio sink and selecting signal profiles based at least in part on that audio context. The signal profiles may include noise cancellation which is configured to facilitate operation within the audio context. Audio context may include user-to-user and user-to-device communications.

Abstract: A method at an electronic device with one or more microphones and a speaker includes receiving a first voice input; comparing the first voice input to one or more voice models; based on the comparing, determining whether the first voice input corresponds to any of a plurality of occupants, and according to the determination, authenticating an occupant and presenting a response, or restricting functionality of the electronic device.

Abstract: Embodiments of the present systems and methods may provide techniques for augmenting textual data that may be used for textual classification tasks. Embodiments of such techniques may provide the capability to synthesize labeled data to improve text classification tasks. Embodiments may be specifically useful when only a small amount of data is available, and provide improved performance in such cases. For example, in an embodiment, a method implemented in a computer system may comprise a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor, and the method may comprise fine-tuning a language model using a training dataset, synthesizing a plurality of samples using the fine-tuned language model, filtering the plurality of synthesized samples, and generating an augmented training dataset comprising the training dataset and the filtered plurality of synthesized sentences.

Abstract: Method and Apparatus for Triggering the Automatic Generation of Narratives Artificial intelligence methods and systems for triggering the generation of narratives are disclosed. Specific embodiments relate to real-time evaluation and automated generation of narrative stories based on received data. For example, data can be tested against data representative of a plurality of story angles to determine whether a narrative story incorporating one or more such story angles is to be automatically generated.

Abstract: A system including at least one computerized device with voice command capability processed remotely includes a low power processor, executing a loose algorithmic model to recognize a wake word prefix in a voice command, the loose model having a low false rejection rate but suffering a high false acceptance rate, and a second processor which can operate in at least a low power/low clock rate mode and a high power/high clock rate mode. When the first processor determines the presence of the wake word, it causes the second processor to switch to the high power/high clock rate mode and to execute a tight algorithmic model to verify the presence of the wake word. By using the two processors in this manner, the average overall power required by the computerized device is reduced, as is the amount of waste heat generated by the system.

Abstract: Techniques facilitating maintenance of tribal knowledge for accelerated compliance control deployment are provided. In one example, a system includes a memory that stores computer executable components and a processor that executes computer executable components stored in the memory, wherein the computer executable components include a knowledge base generation component that generates a knowledge graph corresponding to respective commitments created via tribal exchanges, the knowledge graph comprising a semantic level and an operational level; a semantic graph population component that populates the semantic level of the knowledge graph based on identified parties to the respective commitments; and an operational graph population component that populates the operational level of the knowledge graph based on tracked status changes associated with the respective commitments.

Abstract: Systems and methods include receiving textual data and a predicted answer to a question associated with a text object. The text object includes a structured data field of the textual data. The predicted answer includes a confidence level. The confidence level is determined by a machine learning system based at least in part on one or more models of the machine learning system and the textual data. In response to determining the confidence level being larger than or equal to a predetermined confidence threshold, the predicted answer and a reference is stored in a storage for retrieval and display. The reference indicates a location of the text object in the textual data. In response to determining the confidence level being smaller than the predetermined confidence threshold, the question and the text object associated with the question is displayed, at the user interface, to a user for inputting a true answer.

Abstract: Training a machine learning model such as a neural network, which can automatically extract a hypernym from unstructured data, is disclosed. A preliminary candidate list of hyponym-hypernym pairs can be parsed from the corpus. A preliminary super-term—sub-term glossary can be generated from the corpus, the preliminary super-term—sub-term glossary containing one or more super-term—sub-term pairs. A super-term—sub-term pair can be filtered from the preliminary super-term—sub-term glossary, responsive to detecting that the super-term—sub-term pair is not a candidate for hyponym-hypernym pair, to generate a final super-term—sub-term glossary. The preliminary candidate list of hyponym-hypernym pairs and the final super-term—sub-term glossary can be combined to generate a final list of hyponym-hypernym pairs. An artificial neural network can be trained using the final list of hyponym-hypernym pairs as a training data set, the artificial neural network trained to identify a hypernym given new text data.

Abstract: An urgency level estimation technique of estimating an urgency level of a speaker for free uttered speech, which does not require a specific word, is provided. An urgency level estimation apparatus includes a feature amount extracting part configured to extract a feature amount of an utterance from uttered speech, and an urgency level estimating part configured to estimate an urgency level of a speaker of the uttered speech from the feature amount based on a relationship between a feature amount extracted from uttered speech and an urgency level of a speaker of the uttered speech, the relationship being determined in advance, and the feature amount includes at least one of a feature indicating speaking speed of the uttered speech, a feature indicating voice pitch of the uttered speech and a feature indicating a power level of the uttered speech.

Abstract: Disclosed are a method, device and readable storage medium for speech recognition. The method includes: determining speech features of the speech data by feature extraction on the speech data; determining syllable data corresponding to each of the speech features based on a plurality of feature extraction layers and a softmax function layer included in an acoustic model, where the acoustic model is configured to convert the speech feature into the syllable data; determining text data corresponding to the speech data based on a language model, a pronouncing dictionary and the syllable data, where the pronouncing dictionary is configured to convert the syllable data into the text data, and the language model is configured to evaluate the text data; and outputting the text data.

Abstract: Aspects of the present disclosure include determining, by a processor, an ontology, the ontology comprising a plurality of ontological relationships, receiving, by the processor, a plurality of passages, determining, by the processor, a target set of co-occurring entities comprising a first entity and a second entity, determining a first passage in the plurality of passages that includes the first entity and the second entity, determining, from the ontology, a first ontological relationship between the first entity and the second entity, analyzing the first passage to determine a congruency score for the first ontological relationship, and generating a relationship annotation between the first entity and the second entity in the first passages based on the congruency score being within a threshold.

Abstract: A response method and an apparatus thereof are provided. The method includes: receiving voice information sent through a client by a user; determining, based on the voice information, a user category to which the user belongs; and using voice matching the user category to respond to the voice information. Accordingly, response voice matches the user category of the user, which implements that the response is performed using the response voice targeted for the user category, and thus, the user experience may be improved.

Abstract: A method for instantiating a convolutional neural network on a computing system. The convolutional neural network includes a plurality of layers, and instantiating the convolutional neural network includes training the convolutional neural network using a first loss function until a first classification accuracy is reached, clustering a set of F×K kernels of the first layer into a set of C clusters, training the convolutional neural network using a second loss function until a second classification accuracy is reached, creating a dictionary which maps each of a number of centroids to a corresponding centroid identifier, quantizing and compressing F filters of the first layer, storing F quantized and compressed filters of the first layer in a memory of the computing system, storing F biases of the first layer in the memory, and classifying data received by the convolutional neural network.

Abstract: A method for coordinating actions between an audio channel and a synchronized non-audio channel includes receiving an indication of a start of a session associated with a user and having an audio channel that is synchronized with a non-audio channel. Thereafter, repeated determinations are made as to whether a prompt on the non-audio channel has been received from the user. In response to each determination that the prompt on the non-audio channel has not been received from the user, a signal is sent to cause an inaudible output on the audio channel to the user. In response to a determination that the prompt on the non-audio channel has been received from the user, an audible output is selected based on an activity by the user on the non-audio channel, and a signal is sent to cause the audible output to be output on the audio channel.