Abstract: Term ambiguity is resolved by referencing a terminology database. An input is received comprising the term designated as ambiguous, and a string including the term. The term is posed as a query to the terminology database containing metadata of at least one type. Query results are returned including at least two possible meanings. Sequence(s) are extracted from the query results, each sequence including at least two pieces of metadata of a same type—one for each possible meaning of the ambiguous term. The metadata of each entry of a sequence is compared with the query result and corresponding scores are calculated. The scores are compared to determine a final meaning of the ambiguous term. Simpler embodiments considering one type of metadata (one sequence), may calculate and compare a listing of scores. Complex embodiments considering more than one type of metadata (multiple sequences), may calculate and compare a matrix of scores.

Abstract: A method for converting speech in one of a plurality of input languages into text using machine transliteration and transfer learning is disclosed. The method includes a training stage. The training stage includes receiving a training set of a plurality of audio files and an input text corresponding to the audio input in any input language using the speech recognition engine; transliterating the training set to transform the input text into transliterated text that includes characters of a base language and training acoustic model with the plurality of audio files and corresponding transliterated text using transfer learning. The method further includes an inference stage. The inference stage includes performing decoding on output of the trained acoustic model to generate text includes characters of the base language at inference and transliterating the generated text to output text includes characters in input language using reverse transliteration.

Abstract: In an approach for assisting a user to remain focused on a task, a processor collects a plurality of information regarding one or more digital activities of a user. A processor tokenizes the plurality of information into one or more vectorized embeddings. Responsive to determining that the user has engaged a focus mode, a processor intercepts one or more communications. A processor synthesizes, utilizing a large language model and the one or more vectorized embeddings, one or more natural language responses to the one or more communications. A processor transmits the one or more natural language responses to one or more originating users of the one or more communications. Responsive to determining that the user has ended the focus mode, a processor displays a summary of the one or more communications and the one or more natural language responses transmitted while the user was in the focus mode.

Abstract: A natural language understanding (NLU) framework includes a domain-aware vector encoding (DAVE) framework. The DAVE framework enables a designer to create a DAVE system having a domain-agnostic semantic (DAS) model and a corresponding trained vector translator (VT) model. The DAVE system uses the DAS model to generate domain-agnostic semantic vectors for portions of a user utterance, and then uses the VT model to translate the domain-agnostic semantic vectors into a domain-aware semantic vectors to be used by a NLU system of the NLU framework during a meaning search operation. The VT model is also designed to provide predicted intent classifications for the portions the user utterance. Both the NLU system and the DAVE system of the NLU framework are highly configurable and refer to various NLU constraints during operation, including performance constraints and resource constraints provided by a designer or user of the NLU framework.

Abstract: An artificial intelligence-based semantic recognition method, apparatus, and device. In the artificial intelligence-based semantic recognition method, a pre-trained semantic recognition model is trained by using a training corpus configured by a developer on a model training platform such as a Bot platform and a negative corpus provided on the model training platform, where the negative corpus is extracted by mapping an encoding value of the training corpus to a negative corpus set. Therefore, the negative corpus is extracted based on the encoding value of the training corpus, and a randomized method for generating the negative corpus is changed into a stable method.

Abstract: The implementations herein disclose advanced systems and methods for integrating, analyzing, and reasoning over heterogeneous data at scale. In some implementations, the system comprises a synergistic data processing infrastructure featuring: a graph database core for unified data representation; specialized loaders for concurrent ingestion and processing of structured, unstructured, and time series data; a natural language reasoning engine leveraging large language models; and a multi-modal user interface.

Abstract: The present disclosure provides a method and apparatus for training a neural network, and a method and apparatus for audio processing. The method includes: encoding training audio data input to an encoder network to obtain a first encoding result, and predicting a text label corresponding to the training audio data input to a prediction network to obtain a first prediction result; jointing the first encoding result with the first prediction result to obtain a first joint result; pruning the first encoding result and the first prediction result according to the first joint result to obtain a second encoding result and a second prediction result; performing a joint processing on the second encoding result and the second prediction result input to a joiner network to obtain a second joint result, and adjusting network parameters of the encoder network, the prediction network and the joiner network according to the second joint result.

Abstract: The present invention enables, according to a situation, accurate extraction of related expressions pertaining to search queries and question sentences. A related expression extraction device 1 receives input of text data, performs at least one of categorization of the received text data and determination of a structural pattern of the text data, determines, based on a result of at least one of the categorization of the text data and the determination of the structural pattern of the text data, which of a plurality of comparative assessment models 27 and 28 is used to extract related expression group data 26, and extracts a related expression related to content of the text data from the related expression group data 26 using the determined comparative assessment models 27 and 28.

Abstract: Methods, systems, and computer-readable media for bi-modal generation of natural language (NL) and artificial neural network architectures (NA), with reference to an example implementation framework entitled “ArchGenBERT”. A model and method of training the model for bi-modal generation of NL and NA are described. The model trained for bi-modal generation of NL and NA can be deployed to perform a number of useful tasks to assist with designing, describing, translating, and modifying neural network architectures.

Abstract: A method for execution by a computing device includes identifying sets of identigens for words of a phrase. The method further includes selecting, based on identigen pairing rules of a knowledge database, a first identigen of a first set of identigens for a first word that pairs with at least one corresponding sequentially adjacent identigen of a second set of identigens. The method further includes selecting the first identigen of the first set of identigens as a subsequent identigen corresponding to a subsequent occurrence of the first word within the sequence of words. The method further includes interpreting, in accordance with the identigen pairing rules and based on the first identigen selection for the subsequent occurrence of the first word, the sets of identigens to produce an entigen group that represents a most likely meaning interpretation of the phrase.

Abstract: New question and answer (QA) pairs can be automatically discovered from a corpus of data such as online chats and conversations. Newly discovered QA pairs can augment QA database, which can be used by a computer processor or device, e.g., by a chatbot, an automated machine, and/or another. Existing QA knowledge can be used to learn the structures of QA knowledge distribution in conversations, and new QA knowledge can be automatically learned through the structure of learned QA knowledge distribution in conversations. The structure of learned QA knowledge distribution can be refined by adding more semantics based on labeled data.

Abstract: A method, computer program, and computer system is provided for identifying a speaker in at text based work. Labeled and unlabeled instances corresponding to one or more speakers are extracted. Pseudo-labels are inferred for the extracted unlabeled instances based on the labeled instances. One or more of the unlabeled instances are labeled based on the inferred pseudo-labels.

Abstract: Embodiments described herein provide systems and methods for document recommendation. A system receives a set of training data including a plurality of documents. The system determines whether the set of training data includes annotated contextual information corresponding to the plurality of documents. The system trains supervised and/or unsupervised models based on the availability of data. The models are used to generate vectors representing the documents. During a live text conversation, text from the conversation may be vectorized using the models and the vectors compared to those representing the documents in order to find the most relevant documents. The system may generate an indication of a recommended document.

Abstract: A method, computer program product, and computing system for generating a plurality of acoustic relative transfer functions associated with a plurality of audio acquisition devices of an audio recording system deployed in an acoustic environment. Acoustic relative transfer functions of at least a pair of audio acquisition devices of the plurality of audio acquisition devices may be compared. Location information associated with an acoustic source within the acoustic environment may be determined based upon, at least in part, the comparison of the acoustic relative transfer functions of the at least a pair of audio acquisition devices of the plurality of audio acquisition devices.

Abstract: Systems, devices, and methods related to a Deep Learning Accelerator and memory are described. For example, a microphone may be configured to execute instructions with matrix operands and configured with: a transducer to convert sound waves to electrical signals; an analog to digital converter to generate audio data according to the electrical signals; random access memory to store instructions executable by the Deep Learning Accelerator and store matrices of an Artificial Neural Network; and a controller to store the audio data in the random access memory as an input to the Artificial Neural Network. The Deep Learning Accelerator can execute the instructions to generate an output of the Artificial Neural Network, which may be provided as the primary output of the microphone to a computer system, such as a voice-based digital assistant.

Abstract: Presented herein are systems and methods are presented for detecting out-of-vocabulary (OOV) words in an automatic speech recognition (ASR) system, determining an intended word for the OOV, and adding the intended word to a repository of words. A method may involve receiving audio input data including a series of spoken words; determining that one of the spoken words is an out of vocabulary word absent from a repository of words; generating word candidates based on characteristics of the out of vocabulary word; presenting the word candidates on a display; receiving intended word input data that indicates a selection of one of the word candidates as an intended word for the out of vocabulary word; and adding the intended word to the repository of words. Additionally, one or more devices or apparatuses may be configured to perform such method.

Abstract: The invention provides an efficient implementation of cross-product enhanced high-frequency reconstruction (HFR), wherein a new component at frequency Q?+r?0 is generated on the basis of existing components at ? and ?+?0. The invention provides a block-based harmonic transposition, wherein a time block of complex subband samples is processed with a common phase modification. Superposition of several modified samples has the net effect of limiting undesirable intermodulation products, thereby enabling a coarser frequency resolution and/or lower degree of oversampling to be used. In one embodiment, the invention further includes a window function suitable for use with block-based cross-product enhanced HFR. A hardware embodiment of the invention may include an analysis filter bank, a subband processing unit configurable by control data and a synthesis filter bank.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer-readable media, for speech recognition using multi-dialect and multilingual models. In some implementations, audio data indicating audio characteristics of an utterance is received. Input features determined based on the audio data are provided to a speech recognition model that has been trained to output score indicating the likelihood of linguistic units for each of multiple different language or dialects. The speech recognition model can be one that has been trained using cluster adaptive training. Output that the speech recognition model generated in response to receiving the input features determined based on the audio data is received. A transcription of the utterance generated based on the output of the speech recognition model is provided.

Abstract: Methods and apparatuses for performing context completion to messages in a session are provided in the present disclosure. A message may be obtained. It may be detected that there exists context ellipsis in the message. It may be determined whether the message is retained in the current domain of the session. In response to determining that the message is retained in the current domain, a complementary text for recovering the context ellipsis may be selected in the current domain. A completed message may be generated based on the message and the complementary text.

Abstract: A component management server computer (“server”) and processing methods are disclosed. In some embodiments, the server is programmed to continuously receive input data regarding what is happening in the physical room from one or more input devices. The server is programmed to then detect an utterance of a spoken word from the input data and generate one or more sound metrics based on the input data. Based on the sound metrics as applied to certain criteria, the server is programmed to activate a component, such as an input device, variable, software system, or output device, and cause one or more output devices to execute an action that alerts a user of the activated component. The server can also be programmed to turn on, off, up, or down any of the components based on the activated component.