Abstract: Embodiments are directed to organizing conversation information. Two or more machine learning (ML) models and a plurality of sentences provided from a conversation may be employed to generate insight scores for each sentence such that each insight score correlates to a probability that its sentence includes one or more of an action or a question. In response to one or more sentences having insight scores that exceed a threshold value an information score and a definiteness score may be determined for the one or more sentences. And one or more insights associated with the conversation may be generated based on the one or more sentences. A report may be generated that associates the one or more insights with one or more portions of the conversation that include the one or more sentences that are associated with the insights.

Abstract: In one embodiment, a method includes receiving a user input comprising a natural-language utterance by an assistant xbot from a client system associated with a user, determining a semantic representation of the user input based on a structural ontology defining a labeling syntax for parsing the natural-language utterance to semantic units comprising actions, objects, and attributes, wherein the semantic representation embeds at least one object within at least one action and declares at least one attribute of the embedded object to be acted upon, sending a request based on the semantic representation to an agent for executing a task corresponding to the user input, receiving results of the executed task mapped to a structure determined by the structural ontology from the agent, and sending from the assistant xbot to the client system instructions for presenting a response based on the results of the executed task.

Abstract: An apparatus for decoding data segments representing a time-domain data stream, a data segment being encoded in the time domain or in the frequency domain, a data segment being encoded in the frequency domain having successive blocks of data representing successive and overlapping blocks of time-domain data samples. The apparatus includes a time-domain decoder for decoding a data segment being encoded in the time domain and a processor for processing the data segment being encoded in the frequency domain and output data of the time-domain decoder to obtain overlapping time-domain data blocks. The apparatus further includes an overlap/add-combiner for combining the overlapping time-domain data blocks to obtain a decoded data segment of the time-domain data stream.

Abstract: Systems and processes for providing user-specific acoustic models are provided. In accordance with one example, a method includes, at an electronic device having one or more processors, receiving a plurality of speech inputs, each of the speech inputs associated with a same user of the electronic device; providing each of the plurality of speech inputs to a user-independent acoustic model, the user-independent acoustic model providing a plurality of speech results based on the plurality of speech inputs; initiating a user-specific acoustic model on the electronic device; and adjusting the user-specific acoustic model based on the plurality of speech inputs and the plurality of speech results.

Abstract: Methods, apparatuses, and computer program products are described herein that are configured to express a time in an output text. In some example embodiments, a method is provided that comprises identifying a time period to be described linguistically in an output text. The method of this embodiment may also include identifying a communicative context for the output text. The method of this embodiment may also include determining one or more temporal reference frames that are applicable to the time period and a domain defined by the communicative context. The method of this embodiment may also include generating a phrase specification that linguistically describes the time period based on the descriptor that is defined by a temporal reference frame of the one or more temporal reference frames. In some examples, the descriptor specifies a time window that is inclusive of at least a portion of the time period to be described linguistically.

Abstract: Systems, methods, devices, instructions, and other examples are described for natural language processing. One example includes accessing natural language processing general encoder data, where the encoder data is generated from a general-domain dataset that is not domain specific. A domain specific dataset is accessed and filtered encoder data using a subset of the encoder data is generated. The filtered encoder data is trained using the domain specific dataset to generate distilled encoder data, and tuning values for the distilled encoder data are generated to configure task outputs associated with the domain specific dataset.

Abstract: The invention provides an audio encoder including a combination of a linear predictive coding filter having a plurality of linear predictive coding coefficients and a time-frequency converter, wherein the combination is configured to filter and to convert a frame of the audio signal into a frequency domain in order to output a spectrum based on the frame and on the linear predictive coding coefficients; a low frequency emphasizer configured to calculate a processed spectrum based on the spectrum, wherein spectral lines of the processed spectrum representing a lower frequency than a reference spectral line are emphasized; and a control device configured to control the calculation of the processed spectrum by the low frequency emphasizer depending on the linear predictive coding coefficients of the linear predictive coding filter.

Abstract: A method for compressing a neural network model includes: obtaining a first trained teacher model and a second trained teacher model based on N training samples, N being a positive integer greater than 1; for each of the N training samples, determining a first guide component of the first teacher model and a second guide component of the second teacher model respectively, determining a sub optimization target corresponding to the training sample and configured to optimize a student model according to the first guide component and the second guide component, and determining a joint optimization target based on each of the N training samples and a sub optimization target corresponding to the training sample; and training the student model based on the joint optimization target.

Abstract: Systems and methods to group terms based on context to facilitate determining intent of a command are disclosed. Exemplary implementations to train a model: obtain a set of writings within a particular knowledge domain; obtain a vector generation model that generates vectors for individual instances of the terms in the set of writings; generate a first set of vectors that represent the instances of a first term and other vectors that represent instances of the other terms of the set of writings; train the vector generation model to group the vectors of a similar context in a space of a vector space; obtain a transcript include a new term generated from user audio dictation; generate a new vector that represent the instance of the new term; obtain the space; compare the new vector with the space; utilize the new term as the first term.

Abstract: A machine reading comprehension (MRC) question and answer providing method includes receiving a user question; analyzing the user question; selecting at least one document from at least one domain corresponding to an analyzed user question and searching for a passage, which is a candidate answer determined as being suitable for the user question, in the selected at least one document; obtaining at least one correct answer candidate value by inputting the user question and a corresponding passage into each of at least one MRC question and answer unit; and determining whether the at least one correct answer candidate value is a best answer.

Abstract: A system performs conversations with users using chatbots customized for performing a set of tasks. The system may be a multi-tenant system that allows customization of the chatbots for each tenant. The system processes sentences that may include negation or coreferences. The system determines a confidence score for an input sentence using an intent detection model, for example, a neural network. The system modifies the sentence to generate a modified sentence, for example, by removing a negation or by replacing a pronoun with an entity. The system generates a confidence score for the modified sentence using the intent detection model. The system determines the intent of the sentence based on the confidence scores of the sentence and the modified sentence. The system performs tasks based on the determined intent and performs conversations with users based on the tasks.

Abstract: Systems, computer program products, and methods are described herein for intelligent multimodal classification in a distributed technical environment. The present invention is configured to retrieve one or more multimodal communications from a data repository; initiate one or more feature extraction algorithms on the one or more communication modalities to extract one or more features; generate a training dataset based on at least the one or more features extracted from the one or more communication modalities; initiate one or more machine learning algorithms on the training dataset to generate a first set of parameters; receive an unseen multimodal communication; generate an unseen dataset based on at least the unseen multimodal communication; classify, using the first set of parameters, the unseen multimodal communication into one or more class labels; and initiate an execution of one or more actions on the unseen multimodal communication based on at least the classification.

Abstract: Root cause estimation for a data set corresponding to customer returns of a product may use a probabilistic model to associate customer-entered product return data with probability distributions relating to possible root causes for the returns. A particular application relates to applying a Bayesian network to customer-selected return reason codes and customer-entered return reason comments to estimate a probability distribution for root causes of a plurality of returns and uncertainties relating to the probability distribution estimation. A bag-of-n-grams can be used to enable the Bayesian network to process natural language portions of the customer-entered product return data. The output of the model and other data relating to the root cause estimation can be conveyed to a seller of the returned products via a user interface.

Abstract: An example method for identifying a reading location in a text source as a user reads the text source aloud includes determining phoneme data of the text source, the text source comprising a sequence of words; receiving audio data comprising a spoken word associated with the text source; comparing, by a processing device, the phoneme data of the text source and phoneme data of the audio data; and identifying a location in the sequence of words based on the comparing phoneme data.

Abstract: A computer-implemented method for determining an answer to a question in a multi-party conversation includes receiving a multi-party conversation having multiple nodes of unstructured natural language. Each node is parsed into a plurality of elements. Each element of the plurality of elements that comprises a question is identified. A conversation node list is constructed that identifies relationships between the nodes. At least one answer to the question is produced based on the conversation node list.

Abstract: A question-and-answer system directed to a specific domain optimally utilizes reference documents that are semantically complete for that domain. Semantic completeness of a document is assessed using quality control questions (provided by subject matter experts) applied to the Q&A system followed by analysis of the proposed answers. That analysis is carried out using a cogency module having a feedforward neural network which receives metadata features of the document such as document ownership, document priority, and document type. A domain-optimized corpus for the Q&A system is built by so assessing multiple documents in a document collection, and adding each reference document that is reported as being semantically complete to the domain-optimized corpus. Thereafter, the deep learning question-and-answer system can receive a natural language query from a user, find a responsive answer in the documents while applying the domain-optimized corpus, and provide that answer to the user.

Abstract: Methods, computer program products, and systems are presented. The method computer program products, and systems can include, for instance: receiving, from a user, voice data defining a candidate directive invoking vocal utterance for invoking a directive to execute a first text based command to perform a first computer function of a computer system, wherein the candidate directive invoking vocal utterance includes at least one word or phrase of the text based command, wherein the computer system is configured to perform the first computer function in response to the first text based command and wherein the computer system is configured to perform a second computer function in response to a second text based command; determining, based on machine logic, whether a word or phrase of the candidate vocal utterance sounds confusingly similar to a speech rendering of a word or phrase defining the second text based command.

Abstract: An exemplary method for using a virtual assistant may include, at an electronic device configured to transmit and receive data, receiving a user request for a service from a virtual assistant; determining at least one task to perform in response to the user request; estimating at least one performance characteristic for completion of the at least one task with the electronic device, based on at least one heuristic; based on the estimating, determining whether to execute the at least one task at the electronic device; in accordance with a determination to execute the at least one task at the electronic device, causing the execution of the at least one task at the electronic device; in accordance with a determination to execute the at least one task outside the electronic device: generating executable code for carrying out the least one task; and transmitting the executable code from the electronic device.

Abstract: Exemplary embodiments relate to methods, mediums, and systems for moving language models from a server to the client device. Such embodiments may be deployed in an environment where the server is not able to provide modeling services to the clients, such as an end-to-end encrypted (E2EE) environment. Several different techniques are described to address issues of size and complexity reduction, model architecture optimization, model training, battery power reduction, and latency reduction.

Abstract: An approach for enhancing speech reproduction based on recognizing text segments from sound segments of an audio signal. Enhanced text segments are generated from any text segments whose quality indicators do not reach a threshold level, and enhanced speech segments are synthetized from the enhanced text segments. An enhanced sound is reproduced comprising enhanced sound segments based on the enhanced speech segments.