Abstract: A method for detecting an audio signal and an apparatus, where the method includes determining an input audio signal as a to-be-determined audio signal, determining an enhanced segmental signal-to-noise ratio (SSNR) of the audio signal, where the enhanced SSNR is greater than a reference SSNR, and comparing the enhanced SSNR with a voice activity detection (VAD) decision threshold to determine whether the audio signal is an active signal. Therefore, the method and the apparatus can accurately distinguish an active voice and an inactive voice.

Abstract: A computer-implemented method, according to one embodiment, includes: receiving a complex audio signal which includes an intended audio signal and at least one interfering audio signal. The complex audio signal is converted into text which represents a plurality of words included in the complex audio signal, and at least some of the text is identified as representing words which correspond to the at least one interfering audio signal. The identified text is discarded, and a remaining portion of the text is evaluated to determine whether the remaining portion of the text represents words which convey the voice-based command at an accuracy that is in a predetermined range. Furthermore, the remaining portion of the text is output in response to determining that the remaining portion of the text represents words which convey the voice-based command at an accuracy that is in the predetermined range.

Abstract: An ontology stores information about a domain of an automatic speech recognition (ASR) application program. The ontology is augmented with information that enables subsequent automatic generation of a speech understanding grammar for use by the ASR application program. The information includes hints about how a human might talk about objects in the domain, such as preludes (phrases that introduce an identification of the object) and postludes (phrases that follow an identification of the object).

Abstract: A technology of accurately coding and decoding coefficients which are convertible into linear prediction coefficients even for a frame in which the spectrum variation is great while suppressing an increase in the code amount as a whole is provided. A coding device includes: a first coding unit that obtains a first code by coding coefficients which are convertible into linear prediction coefficients of more than one order; and a second coding unit that obtains a second code by coding at least quantization errors of the first coding unit if (A?1) an index Q commensurate with how high the peak-to-valley height of a spectral envelope is, the spectral envelope corresponding to the coefficients which are convertible into the linear prediction coefficients of more than one order, is larger than or equal to a predetermined threshold value Th1 and/or (B?1) an index Q? commensurate with how short the peak-to-valley height of the spectral envelope is, is smaller than or equal to a predetermined threshold value Th1?.

Abstract: An entity grammar that specifies a computer conversational agent may be received. User utterances are interpreted based on the entity grammar and prompts for the conversational agent to pose are determined based on the entity grammar. An outcome of the dialog is built by storing words in the user utterances and the prompts that match tokens in the entity grammar. The entity grammar specifies both a dialog flow and data structure of the outcome.

Abstract: A computer-implemented method, computer program product, and computer processing system are provided for word embedding. The method includes receiving, by a processor device, a word embedding matrix. The method further includes generating, by a processor device, an average pooling vector and a max pooling vector, based on the word embedding matrix. The method also includes generating, by the processor device, a prediction by applying a Multi-Layer Perceptron (MLP) to the average pooling vector and the max pooling vector.

Abstract: A system and method for extracting relevant data elements from a file for conversion to a tabular format includes a computing device receiving an XML format file having a loop with nested blocks. Each of the blocks has at least one data element. Features are extracted from each data element. These extracted features are processed using a machine learning algorithm to estimate a column header value for the data elements relative to a data schema. With the data element classified, a configuration file is generated to map the column header value to the data elements of the XML file. The configuration file is used to extract the data elements from the XML file to a tabular format. In the healthcare industry, the system and method may be used to extract relevant health information from a clinical document for conversion to a tabular format.

Abstract: A device and method for improving hearing devices by using computer recognition of words and substituting either computer generated words or pre-recorded words in streaming conversation received from a distant speaker. The system may operate in multiple modes such as a first mode being amplification and conditioning of the voice sounds; a second mode having said microphone pickup up the voice sounds from a speaker, a processor configured to convert voice sounds to discrete words corresponding to words spoken by said speaker, generating a synthesized voice speaking said words and outputting said synthesized voice to said sound reproducing element, which is hearable by the user. Other modes include translation of foreign languages into a user's ear and using a heads up display to project the text version of words which the computer had deciphered or translated. The system may be triggered by eye moment, spoken command, hand movement or similar.

Abstract: One embodiment provides a method comprising answering one or more incoming phone calls received at one or more pre-specified phone numbers utilizing a bot. The bot is configured to engage in a conversation with a caller initiating an incoming phone call utilizing a voice recording that impersonates a human being. The method further comprises recording each conversation the bot engages in, and classifying each recorded conversation as one of poison data or truthful training data based on content of the recorded conversation and one or more learned detection models for detecting poisoned data.

Abstract: The present disclosure is generally related to a data processing system to validate vehicular functions in a voice activated computer network environment. The data processing system can improve the efficiency of the network by discarding action data structures and requests that invalid prior to their transmission across the network. The system can invalidate requests by comparing attributes of a vehicular state to attributes of a request state.

Abstract: A method, computer program product, and a system where a processor(s) continuously obtains, from devices in a group of devices within a defined geographic proximity to each other, processing requests. Each request is a result of a device in the group of devices receiving and interpreting a voice command issued within a geographic area comprising the group of devices. The processor(s) buffers, in a memory resource, a portion of the processing requests obtained within a defined time interval. The processor(s) determines there are duplicate processing requests in the portion. Based on determining there are duplicates, the processor(s) rejects the duplicates. The processor(s) selects a specific device to execute each processing request from the devices where the request and the duplicates of that request originated. The processor(s) utilize the specific device to execute the processing request.

Abstract: Of the four primary approaches to processing language by computer, only the parsing approach considers the semantic and syntactic components from the start. In doing so, however, the required resources expand rapidly as the scope of the language processed increases. And as that scope increases, the performance of parsing systems decreases. A natural language processor uses a tumbling-frequency phrase-chain parser as described herein which circumvents this resource-intensive step in parsing, while quickly and almost effortlessly arriving at higher speeds and greater efficiency in natural-language processing with far more accurate results involving a partitioning dictionary and phrase chains, and, more particularly, to the discovery that a small and finite set of “phrase chains” created using a parsing-based phrase-chain processor accounts for a considerable percentage of human language.

Abstract: An embodiment provides a method, including: receiving, via an audio receiver of an information handling device, user voice input; identifying a first word based on the user voice input; accessing a word association data store; selecting an equivalent based on an association with the first word within the word association data store; committing an action based on the equivalent; receiving feedback input from the user regarding the equivalent; and updating the selecting based on the feedback. Other aspects are described and claimed.

Abstract: A method can include receiving a string of characters. The method can include determining one or more possible word boundaries for words in the string of characters based at least partially on a segmentation process. The method can also include determining, for each character in the string of characters, an amount of time between entry of each character on an input device. The method can include determining, based at least partially on the amount of time and the one or more possible word boundaries, one or more actual word boundaries for the words in the string of characters. The method can also include outputting one or more determined words in the string of characters based at least partially on the one or more actual word boundaries.

Abstract: Classification training systems and methods include a neural network for classification of input data, a training dataset providing segmented labeled training data, and a classification training module operable to train the neural network using the training data. A forward pass processing module is operable to generate neural network outputs for the training data using weights and bias for the neural network, and a backward pass processing module is operable to update the weights and biases in a backward pass, including obtaining Region of Target (ROT) information from the training data, generate a forward-backward masking based on the ROT information, the forward-backward masking placing at least one restriction on a neural network output path, compute modified forward and backward variables based on the neural network outputs and the forward-backward masking, and update the weights and biases.

Abstract: A method of operating an electronic device and an electronic device thereof are provided. The method includes receiving a first voice signal of a first user, authenticating whether the first user has authority to control the electronic device, based on the first voice signal, and determining an instruction corresponding to the first voice signal based on an authentication result and controlling the electronic device according to the instruction. The electronic device includes a receiver configured to receive a first voice signal of a first user and at least one processor configured to authenticate whether the first user has authority to control the electronic device based on the first voice signal, determine an instruction corresponding to the first voice signal, and control the electronic device according to the instruction.

Abstract: A classification training system for binary and multi-class classification comprises a neural network operable to perform classification of input data, a training dataset including pre-segmented, labeled training samples, and a classification training module operable to train the neural network using the training dataset. The classification training module includes a forward pass processing module, and a backward pass processing module. The backward pass processing module is operable to determine whether a current frame is in a region of target (ROT), determine ROT information such as beginning and length of the ROT and update weights and biases using a cross-entropy cost function and connectionist temporal classification cost function. The backward pass module further computes a soft target value using ROT information and computes a signal output error using the soft target value and network output value.

Abstract: A classification system and method for training a neural network includes receiving a stream of segmented, labeled training data having a sequence of frames, computing a stream of input features data for the sequence of frames, and generating neural network outputs for the sequence of frames in a forward pass through the training data and in accordance weights and biases. The weights and biases are updated in a backward pass through the training data, including determining Region of Target (ROT) information from the segmented, labeled training data, computing modified forward and backward variables based on the neural network outputs and the ROT information, deriving a signal error for each frame within the sequence of frames based on the modified forward and backward variables, and updating the weights and biases based on the derived signal error. An adaptive learning module is provided to improve a convergence rate of the neural network.

Abstract: A method and system of controlling a digital assistant with dynamically switchable endpoint devices, comprising: dynamically selecting a respective input endpoint device and a respective controlled device for each of a plurality of voice-based requests from a user to the computing system, including: at a first point in time, acquiring respective instances of a first voice input from a first set of two or more input endpoint devices; obtaining a representative copy of the first voice input based on the respective instances of the first voice input that have been acquired from the first set of two or more input endpoint devices; determining a first actionable intent based on the representative copy of the first voice input; and dispatching a first encoded instruction to a first controlled endpoint device selected from the plurality of controlled endpoint devices in accordance with the first actionable intent.

Abstract: A text input system is described for inputting text to a computing device. The text input system has a memory storing first evidence comprising text selected by a user for input to the computing device in a first attempt by a user to input intended text. The memory stores second evidence comprising either information about text deleted by the user or text selected by the user in a second attempt at inputting the intended text. The text input system has an input model configured to combine at least the first and second evidence to produce combined evidence; and a text predictor configured to take the combined evidence as input and use the combined evidence to compute a plurality of predicted text items for input to the computing device.