Abstract: An even-numbered harmonic adder adds a correction value to a first adjacent sample that is the next sample after a first local minimum sample, and subtracts a correction value from a second adjacent sample that is one sample before a local maximum sample. The even-numbered harmonic adder adds a correction value to a third adjacent sample that is the next sample after the local maximum sample, and subtracts a correction value from a fourth adjacent sample that is one sample before a second local minimum sample. An odd-numbered harmonic adder subtracts a correction value from the first local minimum sample and the second local minimum sample, and adds a correction value to the local maximum sample.

Abstract: Multi-modal natural language processing systems are provided. Some systems are context-aware systems that use multi-modal data to improve the accuracy of natural language understanding as it is applied to spoken language input. Machine learning architectures are provided that jointly model spoken language input (“utterances”) and information displayed on a visual display (“on-screen information”). Such machine learning architectures can improve upon, and solve problems inherent in, existing spoken language understanding systems that operate in multi-modal contexts.

Abstract: A method for semantic recognition includes: in response to performing semantic analysis on information acquired by a terminal, a sentence to be processed is acquired. Word recognition is performed on the sentence to be processed, to obtain a plurality of words and part-of-speech information thereof. A target set update operation is determined with a pre-trained word processing model, according to a word to be processed in the set of words to be processed and part-of-speech information of the word to be processed. If a dependency relationship corresponding to the target set update operation is a first dependency relationship, through each of the plurality of preset set update operations, a respective dependency relationship of the word to be processed and a respective confidence level corresponding to the dependency relationship is determined, and a respective update of the set of words to be processed is performed.

Abstract: A device having a voice-based interface activates or “wakes” when it detects an utterance that includes a wakeword; the device may be installed in a vehicle, such as an automobile. The device may distinguish between different wakewords; a different speech processing system may be associated with each wakeword, and each speech processing engine may have its own speech style and associated applications and functions.

Abstract: A transient detector analyzes a given frame n of the input audio signal to determine, based on audio signal characteristics of the given frame n, a transient hangover indicator for a following frame n+1, and signals the determined transient hangover indicator to an associated audio encoder to enable proper encoding of the following frame n+1.

Abstract: An audio encoder has a first information sink oriented encoding branch such as a spectral domain encoding branch, a second information source or SNR oriented encoding branch such as an LPC-domain encoding branch, and a switch for switching between the first and second encoding branches, the second encoding branch having a converter into a specific domain different from the spectral domain such as an LPC analysis stage generating an excitation signal, and the second encoding branch having a specific domain coding branch such as LPC domain processing branch, and a specific spectral domain coding branch such as LPC spectral domain processing branch, and an additional switch for switching between the specific domain coding branch and the specific spectral domain coding branch. An audio decoder has a first domain decoder, a second domain decoder, and a third domain decoder as well as two cascaded switches for switching between the decoders.

Abstract: An information handling system includes a memory device and a processor. The processor instantiates a knowledge base that includes knowledge base information arranged into a parent image locale (PIL) that includes a parent image that presents a portion of the knowledge base information in a first language. The processor translates the knowledge base information into a second language. In translating, the processor further determines that the knowledge base information has previously been translated into the second language based upon the knowledge base information being further arranged into a child image locale (CIL) that includes a child image that presents the portion of the knowledge base information in the second language, and replaces the child image with the parent image in the CIL when a first date when the parent image was modified is later than a second date when the child image was modified.

Abstract: A controller for classifying devices of a building management system (BMS). The controller may be configured to obtain an entity name for a device, extract a core name from the entity name, compare the core name to candidate core names, determine scores for each comparison, identify a highest score, identify a class of a candidate core name, and classify the device in the class.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for identity management are disclosed. In one aspect, a method includes the actions of receiving, from a first computing device, first audio data that includes representations of one or more words in a first voice. The actions further include generating second audio data that includes representations of the one or more words in a second voice. The actions further include providing, for output to a second computing device, the second audio data.

Abstract: Methods, systems, and computer program products for accelerating inference of transformer-based models are provided herein. A computer-implemented method includes obtaining a machine learning model comprising a plurality of transformer blocks, a task, and a natural language dataset; generating a compressed version of the machine learning model based on the task and the natural language dataset, wherein the generating comprises: obtaining at least one set of tokens, wherein each token in the set corresponds to one of the items in the natural language dataset, identifying and removing one or more redundant output activations of different ones of the plurality of transformer blocks for the at least one set of tokens, and adding one or more input activations corresponding to the one or more removed output activations into the machine learning model at subsequent ones of the plurality of the transformer blocks; and outputting the compressed version of the machine learning model to at least one user.

Abstract: The present invention relates to a system that uses natural language processing (NLP) to read data from a file and analyze the data based on user defined parameters. According to an illustrative embodiment of the present disclosure, a system can process and analyze a data file by finding trending themes across data entries. According to a further illustrative embodiment of the present disclosure, the system can search for reoccurring words/phrases based on Ngrams (i.e. n-grams). The system can be adapted to search for Ngrams of varying length depending on the information sought and can sort the results by Ngram length.

Abstract: Data is received that encapsulates a document of text. The text is then segmented into a plurality of semantically coherent units using a coherence-aware text segmentation (CATS) machine learning model. Data is then provided that characterizes the segmenting. Related apparatus, systems, techniques and articles are also described.

Abstract: An IVR and chatbot, or other system, employing a language model, the language model resulting from a method and computer product encoding the method is available for preparing a domain or subdomain specific glossary. The method included using probabilities, word context, common terminology and different terminology to identify domain and subdomain specific language and a related glossary updated according to the method.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for announcing and detecting automated conversation are disclosed. One of the methods includes initiating, over a natural language communication channel, a conversation with a communication participant using a natural language communication method that includes a dialogue of natural language communications. The communication participant is determined to be automated using a pre-defined adaptive interactive protocol that specifies natural language linguistic transformations defined in a sequence. The conversation can be transitioned to a communication method that is different form the natural language communication method in response to determining that the communication participant is automated.

Abstract: Embodiments relate to a method, a computer program and a system for optimizing execution of natural language to structured query language. The method comprises the steps of—receiving a natural language text input and performing natural language processing on the natural language text input to generate a plurality of encoded question tokens. Next, the method comprises performing natural language processing on a plurality of table schema stored in a database to generate a plurality of encoded table schema tokens for each table schema. Further, the method comprises determining a similarity between the plurality of encoded question tokens and the plurality of encoded table schema tokens for at least two table schemas of the plurality of table schema. Furthermore, the method comprises determining an output table schema from the plurality of table schema based on the similarity and outputting a natural language string based on the output table schema.

Abstract: Methods, an encoder and a decoder are configured for transition between frames with different internal sampling rates. Linear predictive (LP) filter parameters are converted from a sampling rate S1 to a sampling rate S2. A power spectrum of a LP synthesis filter is computed, at the sampling rate S1, using the LP filter parameters. The power spectrum of the LP synthesis filter is modified to convert it from the sampling rate S1 to the sampling rate S2. The modified power spectrum of the LP synthesis filter is inverse transformed to determine autocorrelations of the LP synthesis filter at the sampling rate S2. The autocorrelations are used to compute the LP filter parameters at the sampling rate S2.

Abstract: Claim verification is facilitated by identifying a selection of textual content within a user interface, accessing a claim detection module, inputting the textual content into the claim detection module to detect one or more claims within the selection of textual content, accessing an evidence extraction module, using the evidence extraction module to automatically search one or more reference repositories for one or more related references that have content that is related to the detected one or more claims, automatically determining whether the content in the one or more related references supports or refutes the one or more claims using a claim verification module, and, in response to determining a set of references of the one or more related references supports or refutes the one or more claims, presenting a support indicator or a refute indicator within the user interface in association with the one or more claims.

Abstract: A system and method is provided for providing searchable customer call indexes. Consistent with disclosed embodiments, a system may receive call information associated with telephone conversations between callers and a vendor, the call information including an audio recording or transcript for each telephone conversation. The system may also identify one or more keywords from the audio recordings or transcripts and index the call information into one or more indexes based on the identified keywords. Finally, the system may determine search results responsive to a search query based on the indexing. In some embodiments, changes to customer service may be identified based on the search results.

Abstract: Systems and methods described herein relate to adapting a language model for automatic speech recognition (ASR) for a new set of words. Instead of retraining the ASR models, language models and grammar models, the system only modifies one grammar model and ensures its compatibility with the existing models in the ASR system.

Abstract: Techniques are provided for identifying attributes associated with a neologism or an unknown word or name. Real world characteristics can be predicted for the neologism. Trigrams are identified for an input word and word embedding model vector values are calculated for the identified trigrams and entered into a matrix. Trigrams are identified for nearest names. Classification values are calculated based on the trigrams for the input word and the trigrams from the nearest names and the classification values are entered into the matrix. A convolutional neural network can process the matrix to identify one or more characteristics associated with the neologism.