Abstract: This disclosure proposes systems and methods for leveraging entity-related language models in speech processing. A system can receive audio data corresponding to an utterance and perform automatic speech recognition (ASR) on a first portion of the audio data using a general language model. Based on the results, the system may identify a specific language model for processing a second portion of the audio data. The specific language model may include entities belonging to a common subject or class. The specific language model may, in some cases, provide better results than the general language model. While the general language model may describe a whole sentence, the specific language model may describe only a portion of a sentence. Thus, a top-level model may “activate” the specific language model when it may provide useful results. The resulting data may include results from both the general language model and the specific language model.

Abstract: In one example, a computer-based system determines a relationship between a first job and a second job at one or more companies, by using a title data store, a training module, and a prediction module, wherein the title data store accepts job-related information characterizing at least one job-related position that includes at least one of title, corporate entity, job description, and job-related interest data. The training module accepts input data from the title data store, calculates or generates a set of coefficients and a set of job-related vectors from the input data, and stores the coefficients into a database. The prediction module may accept: a first set of data including at least one of a first title, a first corporate designation data, a second set of data including at least one of a second title and a second corporate designation data, and the coefficients from the training module; and then a similarity between the first set of data and the second set of data may be calculated.

Abstract: A method, an apparatus, and a system for speech recognition are provided. a third-party application corresponding to a speech signal of a user can be determined according to the speech signal and by means of semantic analysis; and third-party application registry information is searched for and a third-party program is started, so that the user does not need to tap the third-party application to start the corresponding program, thereby providing more intelligent service for the user and facilitating use for the user.

Abstract: Methods, computer program products, and systems are presented. The methods include, for instance: obtaining an utterance input from a user agent, and collecting context data of the utterance input. A context tag is generated based on the context data, and one or more ground truth having respective utterance semantically identical to the utterance input is selected. Semantical relationship between the context tag and an intent of the selected ground truth is examined and the selected ground truth is updated with the context tag.

Abstract: A device capable of autonomous motion may move in an environment and may receive audio data from a microphone. A model may be trained to process the audio data to determine mask data, which may be used to mask noise in the audio data. Training data for the model may be normalized before training, and different loss functions may be used for different types of training data.

Abstract: In one aspect, the present disclosure relates to a method which, in one example embodiment, can include receiving text data that includes at least unstructured data and wherein the text data is associated with a plurality of messages communicated between a plurality of entities. The method can also include determining relationships between the entities, based on the text data associated with the plurality of messages, and generating, from a knowledge base assembled from at least the text data, a response to a user interaction representing a query for information that corresponds to at least one of the entities and indicates information on one or more of the determined relationships between the entities. The method can also include detecting a deviation in communication between the entities that indicates unauthorized disclosure of information between a first entity and a second entity.

Abstract: An input method, an input device, and an apparatus for input are provided in the embodiments of the present application. The method specifically includes: receiving an input string having a fast input intent, wherein the fast input intent is used to indicate, according to a shorthand information of a word or a phrase corresponding to the input string, the word or the phrase; obtaining word candidates and/or phrase candidates corresponding to the input string according to a language model, wherein the word candidates and the phrase candidates are respectively complete words and complete phrases corresponding to the input string; presenting word candidates and/or phrase candidates to a user. The embodiments of the present application can not only improve the flexibility and application range of the fast input, but also improve the quality of word candidates and/or phrase candidates, thereby improving input efficiency.

Abstract: An electronic device is provided. The electronic device includes a microphone, a processor operatively connected to the microphone, and a memory operatively connected to the processor, wherein the memory, when executed, stores instructions for causing the processor to receive first speech data through the microphone, recognize a user input to call a voice assistant from the first speech data, convert the user input into a first wakeup score, determine the electronic device as a first reference device at least based on the first wakeup score that exceeds a designated threshold value, configure a first noise reduction space based on location information of the first reference device, determine at least one of one or more electronic devices, located in the first noise reduction space, as a first noise reduction device to perform a noise reduction operation, and control the noise reduction operation of the first noise reduction device.

Abstract: Disclosed of the present application is relation to deep learning based voice data processing. The voice data to be detected is converted into target text data based on a voice recognition model so that the keyword text corresponding to the predetermined target voice keyword can be converted. Then, the data is matched with the target text data to determine whether the voice data to be detected includes the target voice keyword based on the matching result. Thus, because the voice recognition model is obtained by deep learning based on the obtained voice recognition data training set, it can obtain high-precision target text data, thereby improving the accuracy of subsequent matching. The problem of low accuracy of detecting voice data for keyword detection can therefor be solved.

Abstract: A method for logging an item of information relating to a rail vehicle, includes recording a speech input having the item of information, by a user of the rail vehicle and saving the recorded speech input as an audio file. The saved audio file is sent via a wireless communications network to a subscriber, remote from the rail vehicle, of the communications network. A device logs the subscriber, remote from a rail vehicle, of the communications network.

Abstract: A human-computer interaction method can include detecting a voice input, and determining whether a first detected voice includes a wake-up word, the wake-up word being intended to wake up an avatar in a social interaction client; displaying the avatar on a live streaming room interface provided by the social interaction client, in response to determining that the first detected voice includes the wake-up word; continuing to detect a voice input, and determining a recognition result by recognizing a second detected voice; determining a user intention based on the recognition result; and controlling, based on the user intention, the avatar to output feedback information.

Abstract: Many headsets include automatic noise cancellation (ANC) which dramatically reduces perceived background noise and improves user listening experience. Unfortunately, the voice microphones in these devices often capture ambient noise that the headsets output during phone calls or other communication sessions to other users. In response, many headsets and communication devices provide manual muting circuitry, but users frequently forget to turn the muting on and/or off creating further problems as they communicate. To address this, the present inventors devised, among other things, an exemplary headset that detects the absence or presence of user speech, automatically muting and unmuting the voice microphone without user intervention. Some embodiments leverage relationships between feedback and feedforward signals in ANC circuitry to detect user speech, avoiding the addition of extra hardware to the headset.

Abstract: An end-to-end deep-learning-based system that can solve both ASR and TTS problems jointly using unpaired text and audio samples is disclosed herein. An adversarially-trained approach is used to generate a more robust independent TTS neural network and an ASR neural network that can be deployed individually or simultaneously. The process for training the neural networks includes generating an audio sample from a text sample using the TTS neural network, then feeding the generated audio sample into the ASR neural network to regenerate the text. The difference between the regenerated text and the original text is used as a first loss for training the neural networks. A similar process is used for an audio sample. The difference between the regenerated audio and the original audio is used as a second loss. Text and audio discriminators are similarly used on the output of the neural network to generate additional losses for training.

Abstract: The present invention provides a pronunciation error detection apparatus capable of following a text without the need for a correct sentence even when erroneous recognition such as a reading error occurs.

Abstract: Techniques and systems are described for performing semantic text searches. A semantic text-searching solution uses a machine learning system (such as a deep learning system) to determine associations between the semantic meanings of words. These associations are not limited by the spelling, syntax, grammar, or even definition of words. Instead, the associations can be based on the context in which characters, words, and/or phrases are used in relation to one another. In response to detecting a request to locate text within an electronic document associated with a keyword, the semantic text-searching solution can return strings within the document that have matching and/or related semantic meanings or contexts, in addition to exact matches (e.g., string matches) within the document. The semantic text-searching solution can then output an indication of the matching strings.

Abstract: An objective of the present invention is to correct a temporal envelope shape of a decoded signal with a small information volume and to reduce perceptible distortions.

Abstract: One embodiment provides a method that includes obtaining a default language corpus. A second language corpus is obtained based on a second language preference. A first transcription of an utterance is received using the default language corpus and natural language processing (NLP). At least one problem word in the first transcription is determined based on an associated grammatical relevance to neighboring words in the first transcription. Upon determining that a first probability score is below a first threshold, an acoustic lookup is performed for an audible match for the problem word in the first transcription based on an associated acoustical relevance. Upon determining that a second probability score is below a second threshold, it is determined whether a match for the problem word exists in the secondary language corpus. Upon determining that the match exists in the secondary language corpus, a second transcription for the utterance is provided.

Abstract: Some embodiments may obtain a natural language question, determine a context of the natural language question, and generate a first vector based on the natural language question using encoder neural network layers. Some embodiments may access a data table comprising column names, generate vectors based on the column names, and determine attention scores based on the vectors. Some embodiments may update the vectors based on the attention scores, generating a second vector based on the natural language question, determine a set of strings comprising a name of the column names and a database language operator based on the vectors. Some embodiments may determine a values based on the determined database language operator, the name, using a transformer neural network model. Some embodiments may generate a query based on the set of strings and the values.

Abstract: A system includes a memory having instructions therein and at least one processor in communication with the memory. The at least one processor is configured to execute the instructions to communicate, into a user device, a deep neural network comprising a predictive audio spectral mask. The at least one processor is also configured to execute the instructions to: generate data corresponding to ambient sound via a multi-microphone device; separate amplitude data and/or phase data from the data via the deep neural network comprising the predictive audio spectral mask; and determine, via the user device and based on the amplitude data and/or phase data, a location of origin of target speech relative to the user device. The at least one processor is configured to execute the instructions to display, via the user device, the location of origin of the target speech relative to the user device.

Abstract: Techniques are provided for training, by a system operatively coupled to a processor, an attention weighted recurrent neural network encoder-decoder (AWRNNED) using an iterative process based on one or more paragraphs of agent sentences from respective transcripts of one or more conversations between one or more agents and one or more customers, and based on one or more customer response sentences from the respective transcripts, and generating, by the system, one or more groups respectively comprising one or more agent sentences and one or more customer response sentences selected based on attention weights of the AWRNNED.