Abstract: The present disclosure relates to systems and methods to extract customized keywords and their corresponding values occurring in a given natural language text. The desired keyword or keywords may occur in different forms, synonyms, abbreviations, and spellings. The disclosed automatic extraction method captures the meaning and context of the desired keywords by transforming the extraction problem into a question answering problem together with capturing the context to narrow down the answer to a unique value for a given keyword. A trained model on an existing corpus of text is used to get a value as an answer to the question phrased using the keyword. When the answer is ambiguous, a context model that uses conditional random field (CRF) is used to provide a most likely value.

Abstract: Systems and methods for device state reversion are disclosed. For example, a requested and/or scheduled device state change may occur, and prior to the device state change, devices may be queried for their device states. This prior device state data may be saved. A user may provide an undo request and the prior device state data may be utilized along with current device state data to select a device to revert device state on, as well as the device state to revert to. In more complex situations and/or when prior state data is unavailable, machine learning techniques may be utilized to select the target device and device state.

Abstract: Systems and methods are provided for performing automated speech recognition. The systems and methods perform operations comprising: accessing a language model that includes a plurality of n-grams, each of the plurality of n-grams comprising a respective sequence of words and corresponding LM score; selecting a target word to boost in the language model; receiving a boosting factor for the target word; identifying a target n-gram in the language model that includes the target word; identifying a subset of n-grams of the plurality of n-grams that include words in a portion of the target n-gram; and adjusting the LM score of the target n-gram based on the LM scores of the subset of n-grams and the boosting factor.

Abstract: A method executed by a computing device includes determining a set of identigens for each query word of a query to produce sets of identigens, where a set of identigens represents different meanings of a word of the query. The method further includes interpreting the sets of identigens to produce a query entigen group. The method further includes accessing a knowledge database utilizing the query entigen group to recover a preliminary response entigen group. The method further includes modifying an answer breadth level based on a response to the preliminary response entigen group to produce an updated answer breadth level. The method further includes accessing the knowledge database utilizing the query entigen group to recover a secondary response entigen group the updated answer breadth level.

Abstract: Methods and apparatus for automated processing of natural language text is described. Received text can be preprocessed to produce language-space data that includes descriptive data elements for words. Source code that includes linguistic constraints, and that may be written in a programming language that is user-friendly to linguists, can be compiled to produce finite-state transducers and bi-machine transducers that are used by a language-processing virtual machine to process the language-space data. The language-processing virtual machine selects and executes code segments in accordance with path transitions of the transducers when applied on automatons to disambiguate meanings of words in the received text.

Abstract: Interactions between organizations occur through multiple channels such as textual communication (e.g., emails) and voice communication (e.g., telephone conversations). All such interaction data collated together constitutes a large amount of unstructured data. A framework is provided for collating the unstructured interaction data and creating a machine-legible structure from it using machine learning models. The machine learning models may generate a variety of generic as well as business-context-relevant insights, with the usage and application of custom-built machine learning model pipelines that generate an overall business insight record that can then be published back into a customer relationship management (CRM) system. Multiple data types are used for the interactions. For example, a voice call may be recorded and stored as an audio file, whereas an email may be stored as a text file. Multiple such formats may also be used to store interaction data.

Abstract: Systems and methods for speech recognition correction include receiving a voice recognition input from an individual user and using a trained error correction model to add a new alternative result to a results list based on the received voice input processed by a voice recognition system. The error correction model is trained using contextual information corresponding to the individual user. The contextual information comprises a plurality of historical user correction logs, a plurality of personal class definitions, and an application context. A re-ranker re-ranks the results list with the new alternative result and a top result from the re-ranked results list is output.

Abstract: A computer-implemented method can include: an audio input device of a portable electronic device receiving verbal speech input from a user and converting the received verbal speech input into an audio input signal; an online processing module of the portable electronic device performing at least one speech recognition operation on the audio input signal; an offline processing module of the portable electronic device performing at least one speech recognition operation on the audio input signal; an interactive game module of the portable electronic device generating user feedback based on results from the at least one speech recognition operation performed by the online processing module and the at least one speech recognition operation by the offline processing module; and a user interface of the portable electronic device providing the user feedback to the user.

Abstract: A method for speech conversion includes receiving, as input to an encoder of a speech conversion model, an input spectrogram corresponding to an utterance, the encoder including a stack of self-attention blocks. The method further includes generating, as output from the encoder, an encoded spectrogram and receiving, as input to a spectrogram decoder of the speech conversion model, the encoded spectrogram generated as output from the encoder. The method further includes generating, as output from the spectrogram decoder, an output spectrogram corresponding to a synthesized speech representation of the utterance.

Abstract: Systems may perform analyses of claims included in a patent document. The systems may generate one or more search strings from the patent document and provide the one or more search strings to a third-party searching authority. The third-party searching authority may return a collection of documents responsive to the one or more search strings. In particular situations, the systems may re-rank the documents of the collection to provide a patent centric ranking. The systems may also analyze the documents of the collection with respect to the elements of the claims to generate various types of patent infringement and/or invalidity reports.

Abstract: A translation device includes a storage unit configured to store a plurality of pieces of learning data, a normalized sentence learning unit configured to perform learning on the plurality of pieces of learning data by combining original text for learning and a corresponding normalized sentence for learning, a translated sentence learning unit configured to perform learning on the plurality of pieces of learning data by combining the original text for learning and a corresponding translated sentence for learning, and a model generation unit configured to generate one normalization/translation model on the basis of a result of learning by the normalized sentence learning unit and the translated sentence learning unit, in which, on at least a part of the learning data, the translated sentence learning unit performs learning after the normalized sentence learning unit performs learning.

Abstract: Implementations described herein relate to causing certain reasoning with respect to why an automated assistant performed (or did not perform) certain fulfillment and/or alternate fulfillment of an assistant command. For example, implementations can receive user input that includes the assistant command, process the user input to determine data to be utilized in performance of the certain fulfillment or the alternate fulfillment of the assistant command, and cause the automated assistant to utilize the data to perform the certain fulfillment or the alternate fulfillment of the assistant command. In some implementations, output that includes the certain reasoning can be provided for presentation to a user in response to additional user input that requests the certain reasoning. In some implementations, a selectable element can be visually rendered and, when selected by the user, the output that includes the certain reasoning can be provided for presentation to the user.

Abstract: A personalized natural language processing system tokenizes a plurality of sets of raw text data to generate a plurality of sets of tokenized text data for the plurality of users, respectively. The tokenized text data includes a sequence of tokens corresponding to the raw text data, the tokens at least identifying distinct words or portions of words in the raw text. The system appends predetermined user-specific tokens to the sets of tokenized text data from the users, respectively. Each predetermined user-specific token corresponds to one of the users. The system processes the sets of tokenized text data using the NLP model in accordance with the appended predetermined user-specific tokens to predict a personalized classification for the sets of tokenized text data from each of the users, and outputs the personalized classifications of the tokenized text data for each of the users.

Abstract: Introduced here are computer programs and associated computer-implemented techniques for discovering the presence of filler words through tokenization of a transcript derived from audio content. When audio content is obtained by a media production platform, the audio content can be converted into text content as part of a speech-to-text operation. The text content can then be tokenized and labeled using a Natural Language Processing (NLP) library. Tokenizing/labeling may be performed in accordance with a series of rules associated with filler words. At a high level, these rules may examine the text content (and associated tokens/labels) to determine whether patterns, relationships, verbatim, and context indicate that a term is a filler word. Any filler words that are discovered in the text content can be identified as such so that appropriate action(s) can be taken.

Abstract: A method includes receiving audio data corresponding to an utterance spoken by a user and processing, using a first recognition model, the audio data to generate a non-contextual candidate hypothesis as output from the first recognition model. The non-contextual candidate hypothesis has a corresponding likelihood score assigned by the first recognition model. The method also includes generating, using a second recognition model configured to receive personal context information, a contextual candidate hypothesis that includes a personal named entity. The method also includes scoring, based on the personal context information and the corresponding likelihood score assigned to the non-contextual candidate hypothesis, the contextual candidate hypothesis relative to the non-contextual candidate hypotheses.

Abstract: A data processing method is described. The method includes acquiring a to-be-filtered dataset, the to-be-filtered dataset including a plurality of pieces of to-be-filtered source language data; filtering all source language data in the to-be-filtered dataset based on a target data filtering model to obtain target source language data remaining after the filtering, the target data filtering model being obtained through training performed by using a reinforcement learning algorithm; and acquiring markup language data corresponding to the obtained target source language data, and acquiring a machine translation model based on the target source language data and the acquired markup language data. In such a data processing process, a filtering rule in the target data filtering model is automatically learned by a machine in a reinforcement learning process. Apparatus and non-transitory computer-readable storage medium counterpart embodiments are also provided.

Abstract: A chatbot system is configured to execute code to perform determining, by the chatbot system, a classification result for an utterance and one or more anchors each anchor of the one or more anchors corresponding to one or more anchor words of the utterance. For each anchor of the one or more anchors, one or more synthetic utterances are generated, and one or more classification results for the one or more synthetic utterances are determined. A report is generated by the chatbot system comprising a representation of a particular anchor of the one or more anchors, the particular anchor corresponding to a highest confidence value among the one or more anchors. The one or more synthetic utterances may be used to generate a new training dataset for training a machine-learning model. The training dataset may be refined according to a threshold confidence values to filter out datasets for training.

Abstract: An audio signal encoding method and apparatus, and an audio signal decoding method and apparatus are disclosed.

Abstract: A technique is described for performing audio-based patient surveys to obtain patient data for use in a health management platform. A virtual assistant is simulated to participate in a natural language conversation with a patient. As part of the conversation, the virtual assistant generates and outputs an audible natural language message that is then presented via a speaker at a user computing device. The audible natural language message includes a prompt for the user to provide a natural language reply. Reply data indicative of a reply by the user is then received and processed to generate and/or update patient data associated with the patient. The patient data is then applied by the health management platform to assist the patient in managing a chronic health condition.

Abstract: This application discloses an audio signal processing method performed by an electronic device. According to this application, embedding processing is performed on a mixed audio signal by mapping the mixed audio signal to an embedding space, to obtain an embedding feature of the mixed audio signal in the embedding space; and generalized feature extraction is performed on the embedding feature, so that a generalized feature of a target component in the mixed audio signal can be obtained through extraction. The generalized feature of the target component has good generalization capability and expression capability, and can be used for different scenarios. Audio signal processing is performed on the mixed audio signal based on the generalized feature of the target component to obtain information of the audio signal of the target object, thereby improving the robustness and generalization of an audio signal processing process, and improving the accuracy of audio signal processing.