Abstract: An electronic apparatus, method, and computer readable medium are provided. The electronic apparatus includes a communicator, and a controller. The controller, based on a first voice input being received, controls the communicator to receive data including first response information corresponding to the first voice input from a server, and outputs the first response information on a display, and based on a second voice input being received, controls the communicator to receive data including second response information corresponding to the second voice input from the server, and outputs the second response information on the display. Based on whether the second voice input is received within a predetermined time from a time corresponding to the output of the first response information, whether a use of utterance history information is identified, and the second response information is displayed differently based on whether the second voice input is received within the predetermined time.

Abstract: Systems and methods for providing supplemental information with a response to a command are provided herein. In some embodiments, audio data representing a spoken command may be received by a cloud-based information system. A response to the command may be retrieved from a category related to the context of the command. A supplemental information database may also be provided that is pre-populated with supplemental information related to an individual having a registered account on the cloud-based information system. In response to retrieving the response to the command, supplemental information may be selected from the supplemental information database to be appended to the response to the command. A message may then be generated including the response and the supplemental information appended thereto, which in turn may be converted into audio data representing the message, which may be sent to a voice-controlled electronic device of the individual.

Abstract: In an example embodiment, text is received at an ecommerce service from a first user, the text in a first language and pertaining to a first listing on the ecommerce service. Contextual information about the first listing may be retrieved. The text may be translated to a second language. Then, a plurality of text objects, in the second language, similar to the translated text may be located in a database, each of the text objects corresponding to a listing. Then, the plurality of text objects similar to the translated text may be ranked based on a comparison of the contextual information about the first listing and contextual information stored in the database for the listings corresponding to the plurality of text objects similar to the translated text. At least one of the ranked plurality of text objects may then be translated to the first language.

Abstract: A method, system, and computer readable medium for decomposing an audio signal into different isolated sources. The techniques and mechanisms convert an audio signal into K input spectrogram fragments. The fragments are sent into a deep neural network to isolate for different sources. The isolated fragments are then combined to form full isolated source audio signals.

Abstract: A computer-implemented method is provided for translating input text from a source language to a target language including receiving, by an interface, the input text in a source language, and identifying, by a processor coupled to the interface, at least one portion of the input text. The method includes replacing each portion with a corresponding sematic structure to produce at least one semantic structure, and organizing the at least one semantic structure into a semantic tree. The method includes matching a portion of the semantic tree to at least one phrase from a stored phrase bank, and providing one or more versions of the at least one phrase in the source language. The method includes receiving a selected version of the set of versions, translating the selected version from the source language to the target language, and providing the selected version in the target language.

Abstract: Methods and apparatus for identifying a music service based on a user command. A content type is identified from a received user command and a music service is selected that supports the content type. A selected music service can then transmit audio content associated with the content type for playback.

Abstract: Systems and methods for analysis and validation of language models trained using data that is unavailable or inaccessible are provided. One example method includes, at an electronic device with one or more processors and memory, obtaining a first set of data corresponding to one or more tokens predicted based on one or more previous tokens. The method determines a probability that the first set of data corresponds to a prediction generated by a first language model trained using a user privacy preserving training process. In accordance with a determination that the probability is within a predetermined range, the method determines that the one or more tokens correspond to a prediction associated with the user privacy preserving training process and outputs a predicted token sequence including the one or more tokens and the one or more previous tokens.

Abstract: Computer implemented method and a system for auto completion of text based on the context associated with the text. The computer implemented method includes steps of receiving input text, identifying a certain context associated with the input text from multiple predefined contexts, by feeding the input text into a context-prediction component of a machine learning model that predicts the certain context, selecting a certain context-specific component of the machine learning model from multiple context-specific components according to the identified certain context, feeding the input text into the selected context-specific component that outputs autocomplete text associated with the identified certain context. The context-specific components are each trained to generate autocompleted text associated with a respective context pre-defined for the respective context-specific component.

Abstract: Methods and systems for deessing of speech signals are described. A deesser of a speech processing system includes an analyzer configured to receive a full spectral envelope for each time frame of a speech signal presented to the speech processing system, and to analyze the full spectral envelope to identify frequency content for deessing. The deesser also includes a compressor configured to receive results from the analyzer and to spectrally weight the speech signal as a function of results of the analyzer. The analyzer can be configured to calculate a psychoacoustic measure from the full spectral envelope, and may be further configured to detect sibilant sounds of the speech signal using the psychoacoustic measure. The psychoacoustic measure can include, for example, a measure of sharpness, and the analyzer may be further configured to calculate deesser weights based on the measure of sharpness. An example application includes in-car communications.

Abstract: A parameter included in a fundamental frequency pattern of a voice can be estimated from the fundamental frequency pattern with high accuracy and the fundamental frequency pattern of the voice can be reconstructed from the parameter included in the fundamental frequency pattern. A learning unit 30 learns a deep generation model including an encoder which regards a parameter included in a fundamental frequency pattern in a voice signal as a latent variable of the deep generation model and estimates the latent variable from the fundamental frequency pattern in the voice signal on the basis of parallel data of the fundamental frequency pattern in the voice signal and the parameter included in the fundamental frequency pattern in the voice signal, and a decoder which reconstructs the fundamental frequency pattern in the voice signal from the latent variable.

Abstract: Provided is a knowledge distillation technique for training a student language model that, relative to a larger teacher language model, has a significantly smaller vocabulary, lower embedding dimensions, and/or hidden state dimensions. Specifically, aspects of the present disclosure are directed to a dual-training mechanism that trains the teacher and student language models simultaneously to obtain optimal word embeddings for the student vocabulary. In some implementations, this approach can be combined with learning shared projection matrices that transfer layer-wise knowledge from the teacher language model to the student language model. Example experimental results have also demonstrated higher compression efficiency and accuracy when compared with other state-of-the-art compression techniques, including the ability to compress the BERTBASE model by more than 60×, with only a minor drop in downstream task metrics, resulting in a language model with a footprint of under 7 MB.

Abstract: A system for audio-visual multi-speaker speech separation. The system includes a processing circuitry and a memory containing instructions that, when executed by the processing circuitry, configure the system to: receive audio signals captured by at least one microphone; receive video signals captured by at least one camera; and apply audio-visual separation on the received audio signals and video signals to provide isolation of sounds from individual sources, wherein the audio-visual separation is based, in part, on angle positions of at least one speaker relative to the at least one camera. The system provides for reliable speech processing and separation in noisy environments and environments with multiple users.

Abstract: Techniques for ordering the output of notification summaries are described. A system may receive multiple notifications intended for a same user or group of users. In response to receiving a user input requesting output of notifications (or in response to multiple notifications expiring soon), the system may identify multiple notifications intended for the user or group of users. The system generates natural language summaries of the notifications, and orders the natural language summaries based on one or more default ordering rules, one or more user preferences, one or more notification provider preference, and/or user feedback. The system then outputs the ordered natural language summaries to the user.

Abstract: Systems and method of diarization of audio files use an acoustic voiceprint model. A plurality of audio files are analyzed to arrive at an acoustic voiceprint model associated to an identified speaker. Metadata associate with an audio file is used to select an acoustic voiceprint model. The selected acoustic voiceprint model is applied in a diarization to identify audio data of the identified speaker.

Abstract: A voice controlled assistant has a housing to hold one or more microphones, one or more speakers, and various computing components. The housing has an elongated cylindrical body extending along a center axis between a base end and a top end. The microphone(s) are mounted in the top end and the speaker(s) are mounted proximal to the base end. A control knob is rotatably mounted to the top end of the housing to rotate about the center axis. A light indicator is arranged on the control knob to exhibit various appearance states to provide visual feedback with respect to the one or more functions being performed by the assistant. In one case, the light indicator is used to uniquely identify participants involved in a call.

Abstract: A system and method for identifying a user of a device includes comparing audio received by a device with acoustic fingerprint information to identify a user of the device. Image data, video data and other data may also be used in the identification of the user. Once the user is identified, operation of the device may be customized based on the user. Further, once the user is identified, data can be associated with the user, for example, usage data, location data, gender data, age data, dominant hand data of the user, and other data. This data can then be used to further customize the operation of the device to the specific user.

Abstract: Features are extracted from an observed speech signal including at least speech of multiple speakers including a target speaker. A mask is calculated for extracting speech of the target speaker based on the features of the observed speech signal and a speech signal of the target speaker serving as adaptation data of the target speaker. The signal of the speech of the target speaker is calculated from the observed speech signal based on the mask. Speech of the target speaker can be extracted from observed speech that includes speech of multiple speakers.

Abstract: Methods, systems, and media for connecting an IoT device to a call are provided. In some embodiments, a method is provided, the method comprising: establishing, at a first end-point device, a telecommunication channel with a second end-point device; subsequent to establishing the telecommunication channel, and prior to a termination of the telecommunication channel, detecting, using the first end-point device, a voice command that includes a keyword; and in response to detecting the voice command, causing information associated with an IoT device that corresponds to the keyword to be transmitted to the second end-point device.

Abstract: In an approach for training data enhancement for an interactive response system, a processor retrieves a set of training data including a set of intents, a set of entities, and a set of utterances that map to each intent. A processor determines iteratively a root verb among the set of utterances for each intent. A processor to determine a set of new intents based on analysis of the determined root verb by performing a pairwise iteration and similarity score over the set of intents. A processor determines iteratively one or more new entities for each new intent. A processor generates a set of new training data based on the set of new intents and entities.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating discrete latent representations of input audio data. Only the discrete latent representation needs to be transmitted from an encoder system to a decoder system in order for the decoder system to be able to effectively to decode, i.e., reconstruct, the input audio data.