Abstract: A user input in a source language is received. A set of contextual data is received. The user input is encoded into a user input feature vector. The set of contextual data is encoded into a context feature vector. The user input feature vector and the context feature vector are used to generate a fusion vector. An adaptive neural network is trained to identify a second context feature vector, based on the fusion vector. A second user input in the source language is received for translation into a target language. The adaptive neural network is used to determine, based on the second context feature vector, a second user input feature vector. The second user input feature vector is decoded, based on the source language and the target language, into a target language output. A user is notified of the target language output.

Abstract: A speech decoding method is performed by a computer device, the speech including a current audio frame and a previous audio frame. The method includes: obtaining a target token corresponding to a smallest decoding score from a first token list including first tokens obtained by decoding the previous audio frame, each first token including a state pair and a decoding score, the state pair being used for characterizing a correspondence between a first state of the first token in a first decoding network corresponding to a low-order language model and a second state of the first token in a second decoding network corresponding to a differential language model; determining pruning parameters according to the target token and an acoustic vector of the current audio frame when the current audio frame is decoded; and decoding the current audio frame according to the first token list, the pruning parameters, and the acoustic vector.

Abstract: Systems and methods for handling multilingual queries are provided. One example method includes receiving, at a computing device, an input, wherein the input comprises a multi-lingual query comprising at least a first source language and a second source language. The multi-lingual query is translated, word for word, into a destination language to produce a monolingual query, with the word order of the multilingual query and the word order of the monolingual query being the same. The monolingual query is processed using natural language processing to map the mono-lingual query to a natural language query in the destination language.

Abstract: Dynamic interfacing with applications is provided. For example, a system receives a first input audio signal. The system processes, via a natural language processing technique, the first input audio signal to identify an application. The system activates the application for execution on the client computing device. The application declares a function the application is configured to perform. The system modifies the natural language processing technique responsive to the function declared by the application. The system receives a second input audio signal. The system processes, via the modified natural language processing technique, the second input audio signal to detect one or more parameters. The system determines that the one or more parameters are compatible for input into an input field of the application. The system generates an action data structure for the application. The system inputs the action data structure into the application, which executes the action data structure.

Abstract: Methods and servers for preparing a sequence for a machine processing task. The method includes acquiring: (i) a vocabulary storing tokens, (ii) a merge table indicating possible mergers between pairs of tokens, and (iii) a text sequence. For a given word from the sequence, the method includes using the vocabulary for splitting the word into an initial sequence, and iteratively merging tokens of the initial sequence to generate a final sequence for the given word. The iterative merging includes, at a given merging iteration using the merge table for identifying merges between pairs of adjacent tokens in a current sequence of the given merging iteration, excluding at least one of merge based on a pre-determined probability, and using the reduced set merges for generating a new sequence by performing at least one merge. The new sequence is to be used as a current sequence during a next merging iteration.

Abstract: A method for training a non-autoregressive TTS model includes obtaining a sequence representation of an encoded text sequence concatenated with a variational embedding. The method also includes using a duration model network to predict a phoneme duration for each phoneme represented by the encoded text sequence. Based on the predicted phoneme durations, the method also includes learning an interval representation and an auxiliary attention context representation. The method also includes upsampling, using the interval representation and the auxiliary attention context representation, the sequence representation into an upsampled output specifying a number of frames. The method also includes generating, based on the upsampled output, one or more predicted mel-frequency spectrogram sequences for the encoded text sequence.

Abstract: Methods, apparatus, systems, and articles of manufacture to fingerprint an audio signal. An example apparatus disclosed herein includes an audio segmenter to divide an audio signal into a plurality of audio segments, a bin normalizer to normalize the second audio segment to thereby create a first normalized audio segment, a subfingerprint generator to generate a first subfingerprint from the first normalized audio segment, the first subfingerprint including a first portion corresponding to a location of an energy extremum in the normalized second audio segment, a portion strength evaluator to determine a likelihood of the first portion to change, and a portion replacer to, in response to determining the likelihood does not satisfy a threshold, replace the first portion with a second portion to thereby generate a second subfingerprint.

Abstract: A method and device for extracting information from acoustic signals receives acoustic signals by a microphone, processes them in an analog front-end circuit, converts the processed signals from the analog front-end circuit to digital signals by sampling at a rate of less than 1 kHz or more preferably less than 500 kHz; and processes the digital signals by a digital back-end classifier circuit. The analog front-end processing decomposes the received signals into frequency components using a bank of analog N-path bandpass filters having different subband center frequencies.

Abstract: A method for determining synthetic speech includes receiving audio data characterizing speech in audio data obtained by a user device. The method also includes generating, using a trained self-supervised model, a plurality of audio features vectors each representative of audio features of a portion of the audio data. The method also includes generating, using a shallow discriminator model, a score indicating a presence of synthetic speech in the audio data based on the corresponding audio features of each audio feature vector of the plurality of audio feature vectors. The method also includes determining whether the score satisfies a synthetic speech detection threshold. When the score satisfies the synthetic speech detection threshold, the method includes determining that the speech in the audio data obtained by the user device comprises synthetic speech.

Abstract: Implementations described herein receive audio data that captures a spoken utterance, generate, based on processing the audio data, a recognition that corresponds to the spoken utterance, and determine, based on processing the recognition, that the spoken utterance is ambiguous (i.e., is interpretable as requesting performance of a first particular action exclusively and is also interpretable a second particular action exclusively). In response to determining that the spoken utterance is ambiguous, implementations determine to provide an enhanced clarification prompt that renders output that is in addition to natural language. The enhanced clarification prompt solicits further user interface input for disambiguating between the first particular action and the second particular action.

Abstract: Provided are an interactive machine translation method and apparatus, a device, and a medium. The method includes: acquiring a source statement input by a user; translating the source statement into a first target statement; determining whether the user adjusts a first vocabulary in the first target statement; and in response to determining that the user adjusts the first vocabulary in the first target statement, acquiring a second vocabulary for replacing the first vocabulary, and adjusting, based on the second vocabulary, a vocabulary sequence located in a front of the first vocabulary and a vocabulary sequence located behind the first vocabulary in the first target statement to generate a second target statement.

Abstract: A method and apparatus for training models in machine translation, an electronic device and a storage medium are disclosed, which relates to the field of natural language processing technologies and the field of deep learning technologies. An implementation includes mining similar target sentences of a group of samples based on a parallel corpus using a machine translation model and a semantic similarity model, and creating a first training sample set; training the machine translation model with the first training sample set; mining a negative sample of each sample in the group of samples based on the parallel corpus using the machine translation model and the semantic similarity model, and creating a second training sample set; and training the semantic similarity model with the second training sample set.

Abstract: A method of obtaining, by an electronic device, a sentence corresponding to context information, including obtaining first output information including at least one word output by decoding the context information based on at least one data; based on detecting that a first token is not included in the first output information, determining whether a number of words included in the first output information is greater than or equal to a reference value; based on a result of the determining, replacing the at least one data with other data; and obtaining the sentence corresponding to the context information based on at least one output information obtained by decoding the context information based on the other data.

Abstract: Systems, apparatuses, methods, and computer program products are disclosed for determining robustness information for an NLP model. Modification rules, such as replacement rules and/or insertion rules, are used to generate instances of modified test data based on instances of test data that comprise words and have a syntax and a semantic meaning. The instances of test data and modified test data are provided to the NLP model and the output of the NLP model is analyzed to determine output changing instances of modified test data, which are instances of modified test data yielded output from the NLP model that is different and/or not similar to the output yielded from the NLP model for the corresponding instance of test data. Robustness information for the NLP model is determined based at least in part on the output changing instances of modified test data. White and/or black box attacks may be performed.

Abstract: Systems and methods for generating a natural language domain corpus to train a machine learning natural language understanding process. A base utterance expressing an intent and an intent profile indicating at least one of categories, keywords, concepts, sentiment, entities, or emotion of the intent are received. Machine translation translates the base utterance into a plurality of foreign language utterances and back into respective utterances in the target natural language to create a normalized utterance set. Analysis of each utterance in the normalized utterance set determines respective meta information for each such utterance. Comparison of the meta information to the intent profile determines a highest ranking matching utterance within the normalized utterance set. A set of natural language data to train a machine learning natural language understating process is created based on further natural language translations of the highest ranking matching utterance.

Abstract: A learning device of a phrase generation model includes a memory; and a processor configured to execute learning the phrase generation model including an encoder and a decoder, by using, as training data, a 3-tuple. The 3-tuple includes a combination of phrases and at least one of a conjunctive expression representing a relationship between the phrases, and a relational label indicating the relationship represented by the conjunctive expression. The encoder is configured to convert a phrase into a vector from a 2-tuple. The 2-tuple includes a phrase and at least one of the conjunctive expression and the relational label. The decoder is configured to generate, from the converted vector and the conjunctive expression or the relational label, a phrase having the relationship represented by the conjunctive expression or the relational label with respect to the phrase.

Abstract: A method for identifying and ranking potentially privileged documents using a machine learning topic model may include receiving a set of documents. The method may also include, for each of two or more documents in the set of documents, extracting a set of spans from the document, generating, using a machine learning topic model, a set of topics and a subset of legal topics for the set of spans, generating a vector of probabilities for each span with a probability being assigned to each topic in the set of topics for the span, assigning a score to one or more spans in the set of spans by summing the probabilities in the vector that are assigned to a topic in the subset of legal topics, and assigning a score to the document. The method may further include ranking the two or more documents by their assigned scores.

Abstract: Example methods and apparatus for providing voice alignment are described. One example method including: obtaining an original voice and a test voice, the test voice is a voice generated after the original voice is transmitted over a communications network; performing loss detection and/or discontinuity detection on the test voice, the loss detection is used to determine whether the test voice has a voice loss compared with the original voice, and the discontinuity detection is used to determine whether the test voice has voice discontinuity compared with the original voice; and aligning the test voice with the original voice based on a result of the loss detection and/or the discontinuity detection, to obtain an aligned original voice and an aligned test voice, the result of the loss detection and/or the discontinuity detection is used to indicate a manner of aligning the test voice with the original voice.

Abstract: In methods for training a natural language generation (NLG) model using a processor a document-level machine translation (MT) model is provided by training an MT model to receive as input, token sequences in a first language, and to generate as output, token sequences in a second language. An augmented document-level MT model is provided by training the document-level MT model to receive as input, paired language-independent structured data and token sequences in the first language, and to generate as output, token sequences in the second language. The augmented document-level MT model is trained to receive as input, language-independent structured data, and to generate as output, token sequences in the second language.

Abstract: An electronic device is provided. The electronic device includes a memory and a processor. The processor is configured to, based on acquiring a first sentence in a first language, determine whether to correct the first sentence to another sentence in the first language by using a second language model trained based on a learning corpus, and based on determining to correct the first sentence to another sentence in the first language, input the first sentence into a conversion model trained to acquire another sentence having a similarity greater than or equal to a threshold value to an input sentence and acquire a second sentence in the first language which is a corrected form of the first sentence, and based on acquiring the second sentence, input the second sentence into a translation model trained based on the learning corpus and acquire a third sentence in a second language.