Abstract: The present invention relates to an apparatus and method for automatically generating machine reading comprehension training data, and more particularly, to an apparatus and method for automatically generating and managing machine reading comprehension training data based on text semantic analysis. The apparatus for automatically generating machine reading comprehension training data according to the present invention includes a domain selection text collection unit configured to collect pieces of text data according to domains and subjects, a paragraph selection unit configured to select a paragraph using the pieces of collected text data and determine whether questions and correct answers are generatable, and a question and correct answer generation unit configured to generate questions and correct answers from the selected paragraph.

Abstract: An end-to-end automatic speech recognition (ASR) system includes: a first encoder configured for close-talk input captured by a close-talk input mechanism; a second encoder configured for far-talk input captured by a far-talk input mechanism; and an encoder selection layer configured to select at least one of the first and second encoders for use in producing ASR output. The selection is made based on at least one of short-time Fourier transform (STFT), Mel-frequency Cepstral Coefficient (MFCC) and filter bank derived from at least one of the close-talk input and the far-talk input. If signals from both the close-talk input mechanism and the far-talk input mechanism are present for a speech segment, the encoder selection layer dynamically selects between the close-talk encoder and the far-talk encoder to select the encoder that better recognizes the speech segment. An encoder-decoder model is used to produce the ASR output.

Abstract: According to one embodiment, a speech recognition apparatus includes processing circuitry. The processing circuitry generates, based on sensor information, environmental information relating to an environment in which the sensor information has been acquired, generates, based on the environmental information and generic speech data, an adapted acoustic model obtained by adapting a base acoustic model to the environment, acquires speech uttered in the environment as input speech data, and subjects the input speech data to a speech recognition process using the adapted acoustic model.

Abstract: Methods and systems are described for generating dynamic conversational responses using machine learning models. The dynamic conversational responses may be generated in real time and reflect the likely goals and/or intents of a user. The machine learning model may provide these features by monitoring one or more user actions and/or lengths of time between one or more user actions during conversational interactions.

Abstract: Mitigating mistranscriptions resolves errors in a transcription of the audio portion of a video based on a semantic matching with contextualized data electronically garnered from one or more sources other than the audio portion of the video. A mistranscription is identified using a pretrained word embedding model that maps words to an embedding space derived from the contextualizing data. A similarity value for each vocabulary word of a multi-word vocabulary of the pretrained word embedding model is determined in relation to the mistranscription. Candidate words are selected based on the similarity values, each indicating a closeness of a corresponding vocabulary word to the mistranscription. The textual rendering is modified by replacing the mistranscription with a candidate word that, based on average semantic similarity values, is more similar to the mistranscription than is each other candidate word.

Abstract: Systems, methods, and apparatuses are described for determining compliance with a plurality of restrictions associated with one or more devices in an organization. First text indicating restrictions may be received, and second text indicating a current configuration of one or more devices may be received. Both sets of text may be processed by, e.g., removing a portion of the text based on a predetermined list of terms and simplifying the text using a lemmatization algorithm. A first vector and second vector may be generated based on the processed sets of text, and each vector may be weighted based on an inverse frequency of words in their respective text. Each vector may be normalized based on semantic analysis. The two vectors may be compared. Based on the comparison, third text corresponding to a portion of the second vector may be generated and transmitted to a third computing device.

Abstract: This application relates to apparatus and methods for natural language understanding in conversational systems using machine learning processes. In some examples, a computing device receives a request that identifies textual data. The computing device applies a natural language model to the textual data to generate first embeddings. In some examples, the natural language model is trained on retail data, such as item descriptions and chat session data. The computing device also applies a dependency based model to the textual data to generate second embeddings. Further, the computing device concatenates the first and second embeddings, and applies an intent and entity classifier to the concatenated embeddings to determine entities, and an intent, for the request. The computing device may generate a response to the request based on the determined intent and entities.

Abstract: According to one embodiment, an information processing apparatus includes following units. The acquisition unit acquires first training data including a combination of a voice feature quantity and a correct phoneme label of the voice feature quantity. The training unit trains an acoustic model using the first training data in a manner to output the correct phoneme label in response to input of the voice feature quantity. The extraction unit extracts from the first training data, second training data including voice feature quantities of at least one of a keyword, a sub-word, a syllable, or a phoneme included in the keyword. The adaptation processing unit adapts the trained acoustic model using the second training data to a keyword detection model.

Abstract: Various embodiments of an apparatus, methods, systems and computer program products described herein are directed to a Voice Analysis Engine. According to various embodiments, the Voice Analysis Engine receives first streaming prompt data from a computing device. The Voice Analysis Engine analyzes the first streaming prompt data to provide feedback to the user of the computing device. Upon determining the first streaming prompt data satisfies one or more criteria, the Voice Analysis Engine receives second streaming prompt data from the computing device. The Voice Analysis Engine analyzes the streaming prompt data to predict a respiratory state of the user of the computing device.

Abstract: A voice recognition method is provided. The voice recognition method includes: collecting a plurality of voice signals; extracting the voiceprint features of each of the voice signals; performing a data process on the voiceprint features, to convert the voiceprint features into a N-dimensional matrix, and N is an integer greater than or equal to 2; performing a feature normalization process on the N-dimensional matrix to obtain a plurality of voiceprint data; classifying the voiceprint data to generate a clustering result; finding out a centroid of each cluster according to the clustering result, and registering the voiceprint data adjacent to each of the centroid. The disclosure also provides an electronic device that adapted for the voice recognition method.

Abstract: An audio processing method includes the following operations. A calculated value is obtained according to multiple audio clock frequency information contained in multiple audio input packets. An audio sampling frequency is generated according to the calculated value and a link symbol clock signal. Multiple audio output packets corresponding to the audio input packets are generated according to the audio sampling frequency.

Abstract: This application discloses an information conversion method and apparatus, a storage medium, and an electronic apparatus.

Abstract: A keyword spotting apparatus, method, and computer-readable recording medium are disclosed. The keyword spotting method using an artificial neural network according to an embodiment of the disclosure may include obtaining an input feature map from an input voice; performing a first convolution operation on the input feature map for each of n different filters having the same channel length as the input feature map, wherein a width of each of the filters is w1 and the width w1 is less than a width of the input feature map; performing a second convolution operation on a result of the first convolution operation for each of different filters having the same channel length as the input feature map; storing a result of the second convolution operation as an output feature map; and extracting a voice keyword by applying the output feature map to a learned machine learning model.

Abstract: There is provided a computer-implemented method of training a speech-to-speech (S2S) machine learning (ML) model for adapting at least one voice attribute of speech, comprising: creating an S2S training dataset of a plurality of S2S records, wherein an S2S record comprises: a first audio content comprising speech having at least one first voice attribute, and a ground truth label of a second audio content comprising speech having at least one second voice attribute, wherein the first audio content and the second audio content have the same lexical content and are time-synchronized, and training the S2S ML model using the S2S training dataset, wherein the S2S ML model is fed an input of a source audio content with at least one source voice attribute and generates an outcome of the source audio content with at least one target voice attribute.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for reducing audio noise are disclosed. In one aspect, a method includes the actions of receiving first audio data of a user utterance. The actions further include determining an energy level of second audio data being outputted by the loudspeaker. The actions further include selecting a model from among (i) a first model that is trained using first audio data samples that each encode speech from one speaker and (ii) a second model that is trained using second audio data samples that each encode speech from either one speaker or two speakers. The actions further include providing the first audio data as an input to the selected model. The actions further include receiving processed first audio data. The actions further include outputting the processed first audio data.

Abstract: An apparatus for encoding a first channel and a second channel of an audio input signal including two or more channels to obtain an encoded audio signal according to an embodiment includes a normalizer configured to determine a normalization value for the audio input signal depending on the first channel of the audio input signal and depending on the second channel of the audio input signal. Moreover, the apparatus includes an encoding unit configured to generate a processed audio signal having a first channel and a second channel. The encoding unit is configured to encode the processed audio signal to obtain the encoded audio signal.

Abstract: Systems, devices, and methods provide improved autonomous agents by creating a concept lattice that represent objects and attributes and using the concept lattice to manage a dialogue with a user device. An autonomous agent application can receive queries from a user and serve response (e.g., responses identifying objects and/or object attributes) based on one or more traversals of a concept lattice. In some embodiments, the concept lattice can be generated from tabular data indicating a set of objects and respective sets of attributes. The user can provide further input to traverse from one selected node to another within the concept lattice in order to identify other nodes in the concept lattice that meet the user's needs.

Abstract: A computer process for entity resolution of natural language records including training a semantic embedding function on a corpus of unlabeled training materials. The semantic embedding function can take a word and represent it as a vector, where the vector represents the word as it relates to the semantic information of the corpus of unlabeled training materials. The process may transform a list of normalized descriptions using the semantic embedding function into a list of vector representations of the descriptions. The process may transform words from a natural language record to a vector representation of the natural language record using the semantic embedding function, and may use a named entity recognizer. The process may find a best match description from the list of normalized descriptions using the list of vector representations of the descriptions and the vector representation of the natural language record, and may include using word mover distance.

Abstract: A hearing aid system presents a hearing impaired user with customized enhanced intelligibility speech sound in a preferred language while maintaining the voice identity of speaker. The system includes a neural network model trained with a set of source speech data representing sampling from a speech population relevant to the user. The model is also custom trained with a set of parallel or non-parallel alternative articulations, collected during an interactive session with user or algorithmically generated based on the hearing profile of the user or category of users with common linguistic and hearing profiles.

Abstract: Various embodiments of a computer-implemented system which learns textual representations while filtering out potentially personally identifying data and retaining semantic meaning within the textual representations are disclosed herein.