Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for selecting units for speech synthesis. One of the methods includes determining a sequence of text units that each represent a respective portion of text for speech synthesis; and determining multiple paths of speech units that each represent the sequence of text units by selecting a first speech unit that includes speech synthesis data representing a first text unit; selecting multiple second speech units including speech synthesis data representing a second text unit based on (i) a join cost to concatenate the second speech unit with a first speech unit and (ii) a target cost indicating a degree that the second speech unit corresponds to the second text unit; and defining paths from the selected first speech unit to each of the multiple second speech units to include in the multiple paths of speech units.

Abstract: The disclosure can provide systems and methods for providing globalization features in a service management application interface. In one embodiment, a method can include receiving a definition comprising at least one function written in a first language; embedding, within the definition, the at least one function written in a second language translatable to the first language; retrieving the at least one function written in the second language translatable to the first language; and based at least in part on the at least one function written in the second language, converting a text string from the first language to the second language, wherein the text string comprises at least a portion of the definition.

Abstract: Methods, systems, and computer program products for unsupervised learning of interpretable conversation models from conversation logs are provided herein. A computer-implemented method includes obtaining human-to-human conversation logs; training a deep learning model by (i) learning, in an unsupervised manner, semantic labels for dialog contexts in the multiple human-to-human conversation logs, (ii) mapping the learned semantic labels to query responses across the multiple human-to-human conversation logs, and (iii) inferring one or more entities from the multiple conversation logs based at least in part on the mapping; constructing a human-interpretable conversation model based at least in part on patterns determined via the trained deep learning model; and outputting the human-interpretable conversation model to at least one user.

Abstract: A computer program product comprises a non-transitory computer readable storage that has a computer readable program stored thereon. When executed on a computer, the computer readable program causes the computer to select, with a processor, a multilingual graphical user interface template corresponding to a three-step skincare treatment process. The multilingual graphical user interface template has a layout of all of the objects present in a multilingual graphical user interface without any content displaying a human-spoken language. Furthermore, the computer is caused to determine, with the processor, a preferred spoken language of a user operating a computing device. Additionally, the computer is caused to automatically select, from a multilingual object database, one or more audiovisual objects composed in the preferred spoken language and one or more menu objects composed in the preferred spoken language.

Abstract: A personal authentication device includes: acoustic signal transmission means 701 for transmitting a first acoustic signal to a part of a head of a user; acoustic signal observation means 702 for observing a second acoustic signal which is an acoustic signal after the first acoustic signal propagates through the part of the head; acoustic property calculation means 703 for calculating an acoustic property from the first acoustic signal and the second acoustic signal; and user identification means 704 for identifying the user, based on the acoustic property or a feature value extracted from the acoustic property and relating to the user.

Abstract: Disclosed are systems, methods, and non-transitory computer-readable media for using a language classification system for generating response messages. A messaging system receives a message transmitted from a first user to a second user, and determines, based on a set of language counter values determined for the second user, a probability value that the second user will respond to the message in a first language and a probability value that the second user will respond to the message in a second language. The language counter values are determined using a text classification model and indicate a number of times that the second user has used the languages in previous messages. Based on the probability values, the messaging system determines that the second user will respond to the message in the first language and causes a set of recommended responses in the first language be presented on the second client device.

Abstract: Methods, apparatuses, and computer program product are described herein configured to transform an input model into a format that can be expressed causally in one or more phrases with a narrative representation. The methods, apparatuses, and computer program product may be configured to obtain the input model representing a data set composed of a plurality of nodes, a plurality of edges, and a plurality of paths and identify a path of maximal significance. The methods, apparatuses, and computer program product include identifying a subset of nodes from the plurality of nodes and a subset of edges from the plurality of edges associated with the path of maximal significance. The methods, apparatuses, and computer program product are configured to generate, using a natural language generation system, one or more phrases with a narrative representation describing the subset of nodes and the subset of edges associated with the path of maximal significance.

Abstract: Techniques are disclosed to optimize feature selection in generating betas for a feature dictionary of a neuro-linguistic Cognitive AI System. A machine learning engine receives a sample vector of input data to be analyzed by the neuro-linguistic Cognitive AI System. The neuro-linguistic Cognitive AI System is configured to generate multiple betas for each of a plurality of sensors. The machine learning engine identifies a sensor specified in the sample vector and selects optimization parameters for generating betas based on the identified sensor.

Abstract: A data processing system analyzes a corpus of conversation data received at an interactive conversation service to train a response recommendation model. The response recommendation model generates response vectors based on custom responses and using the trained model and generates a context vector based on received input at the interactive conversation service. The context vector is compared to the set of response vectors to identify a set of recommended responses, which are recommended to an agent conversing with a user using the interactive conversation service.

Abstract: An apparatus for generating a decoded two-channel signal includes: an audio processor for decoding an encoded two-channel signal to obtain a first set of first spectral portions; a parametric decoder for providing parametric data for a second set of second spectral portions and a two-channel identification identifying either a first or a second different two-channel representation for the second spectral portions; and a frequency regenerator for regenerating a second spectral portion depending on a first spectral portion of the first set of first spectral portions, the parametric data for the second portion and the two-channel identification for the second portion.

Abstract: An echo delay tracking method includes obtaining, by a computing terminal, a current frame reference signal and a current frame collection signal; and identifying target frequency information in the current frame collection signal according to signal-to-noise ratios of the current frame collection signal at a plurality of frequencies. The target frequency information includes at least one target frequency that corresponds to a signal-to-noise ratio greater than a first preset threshold. The method also includes determining whether the current frame collection signal is a valid frame signal based on the target frequency information; and if yes, performing a cross-correlation operation of a signal component of the current frame reference signal corresponding to the target frequency and a signal component of the current frame collection signal corresponding to the target frequency. An echo delay value can be obtained according to a result of the cross-correlation operation.

Abstract: A computer-implemented method, computer program product, and apparatus are provided. The method includes generating a plurality of sequences of small unit tokens from a first language model that is trained with a small unit corpus including the small unit tokens, the small unit corpus having been derived by tokenization with a small unit. The method further includes tokenizing the plurality of sequences of small unit tokens by a large unit that is larger than the small unit, to create a derived large unit corpus including derived large unit tokens.

Abstract: A training method of an acoustic model includes constructing window-level input speech data based on a speech sequence; inputting the window-level input speech data to an acoustic model; calculating a sequence level-error based on an output of the acoustic model; acquiring window-level errors based on the sequence level-error; and updating the acoustic model based on the window-level errors.

Abstract: An electronic apparatus is provided. The electronic apparatus according to an embodiment includes an audio input unit configured to receive sound sources from different positions and generate a plurality of voice signals, a pre-processor configured to perform pre-processing of the plurality of voice signals, and a voice recognition unit configured to perform voice recognition using the plurality of voice signals pre-processed by the pre-processor, and in response to a predetermined trigger being detected as a result of the voice recognition, generate trigger information, wherein the pre-processor is further configured to receive feedback on the trigger information generated by the voice recognition unit, change a pre-processing method according to the trigger information, process the plurality of voice signals using the changed pre-processing method, and generate enhanced voice signals.

Abstract: Embodiments of the present disclosure provide signal encoding and decoding methods and devices. The method includes: determining, a quantity k of subbands to be encoded, where i is a positive number, and k is a positive integer; selecting, according to quantized envelopes of all subbands, k subbands from all the subbands, or selecting k subbands from all subbands according to a psychoacoustic model; and performing a first-time encoding operation on spectral coefficients of the k subbands. In some embodiments of the present disclosure, the quantity k of subbands to be encoded is determined according to the quantity of available bits and the first saturation threshold, and encoding is performed on the k subbands that are selected from all the subbands, instead of on an entire frequency band.

Abstract: Systems and methods for identifying a person's native language, are presented. A native language identification system, comprising a plurality of artificial neural networks, such as time delay deep neural networks, is provided. Respective artificial neural networks of the plurality of artificial neural networks are trained as universal background models, using separate native language and non-native language corpora. The artificial neural networks may be used to perform voice activity detection and to extract sufficient statistics from the respective language corpora. The artificial neural networks may use the sufficient statistics to estimate respective T-matrices, which may in turn be used to extract respective i-vectors. The artificial neural networks may use i-vectors to generate a multilayer perceptron model, which may be used to identify a person's native language, based on an utterance by the person in his or her non-native language.

Abstract: A computer program product for domain-specific data generation. The computer program product including a computer-readable storage medium having program instructions embodied therewith. The program instructions executable by a processor to cause the processor to perform operations including performing an analysis on a domain-specific corpus to identify a base term and a modifier term. The modifier term modifies the base term in at least a portion of the domain-specific corpus. The operations also include accessing a first entry in lexicon data. The first entry including core data corresponding to domain-independent lexical information for the base term. The operations further include adding, based on the analysis, non-core data to the first entry, the non-core data corresponding to domain-specific lexical information for the base term, wherein the non-core data identifies the modifier term as a domain-specific modifier of the base term.

Abstract: A method includes performing, at a device, an analysis of a domain-specific corpus to identify a base term and a modifier term. The modifier term modifies the base term in at least a portion of the domain-specific corpus. The method also includes accessing, by the device, a first entry in lexicon data. The first entry includes core data corresponding to domain-independent lexical information for the base term. The method further includes adding, based on the analysis, non-core data to the first entry. The non-core data corresponds to domain-specific lexical information for the base term. The non-core data identifies the modifier term as a domain-specific modifier of the base term.

Abstract: A voice input by a vehicle user is taken as a basis for determining at least one keyword from a set of prescribed keywords. The at least one keyword is taken as a basis for determining at least one event and/or at least one state from a set of events and/or states of the vehicle that are stored during a prescribed period of time. This involves the respective event and/or the respective state being stored in conjunction with at least one condition occurrence that characterizes a respective condition that needs to be met in order for the event to occur and/or the respective state to exist. In addition, a response is determined from a set of prescribed responses on the basis of the condition occurrence that is associated with the determined event and/or state. Furthermore, a signaling signal is determined on the basis of the ascertained response.

Abstract: Embodiments of the present disclosure provide a decoding method and a decoding apparatus. The decoding method includes: in a case in which it is determined that a current frame is a lost frame, synthesizing a high frequency band signal; determining subframe gains of multiple subframes of the current frame; determining a global gain of the current frame; and adjusting, according to the global gain and the subframe gains of the multiple subframes, the synthesized high frequency band signal to obtain a high frequency band signal of the current frame. A subframe gain of the current frame is obtained according to a gradient between subframe gains of subframes previous to the current frame, so that transition before and after frame loss is more continuous, thereby reducing noise during signal reconstruction, and improving speech quality.