Abstract: Systems and methods are described herein for providing media guidance. Control circuitry may receive a first voice input and access a database of topics to identify a first topic associated with the first voice input. A user interface may generate a first response to the first voice input, and subsequent to generating the first response, the control circuitry may receive a second voice input. The control circuitry may determine a match between the second voice input and an interruption input such as a period of silence or a keyword or a phrase, such as “Ahh,”, “Umm,”, or “Hmm.” The user interface may generate a second response that is associated with a second topic related to the first topic. By interrupting the conversation and changing the subject from time to time, media guidance systems can appear to be more intelligent and human.

Abstract: A method and apparatus for providing a voice recognition service can include separating a user utterance from noise and converting the user utterance into a text to generate content of the user utterance, extracting, from the content of the user utterance, call words information, domain information, end service name information, and operations information, generating a corrected user command by correcting a user command, and generating response information by using the corrected user command.

Abstract: An estimation apparatus includes: an input part configured to receive, as inputs, an extended word alignment between a source sentence and a translated sentence, and translation quality tags between the source sentence and the translated sentence; and an edit tag estimation part configured to estimate an edit tag based on the extended word alignment and the translation quality tags.

Abstract: The various implementations described herein include methods and systems for determining device leadership among voice interface devices. In one aspect, a method is performed at a first electronic device of a plurality of electronic devices, each having microphones, a speaker, processors, and memory storing programs for execution by the processors. The first device detects a voice input. It determines a device state and a relevance of the voice input. It identifies a subset of electronic devices from the plurality to which the voice input is relevant. In accordance with a determination that the subset includes the first device, the first device determines a first score of a criterion associated with the voice input and receives second scores of the criterion from other devices in the subset. In accordance with a determination that the first score is higher than the second scores, the first device responds to the detected input.

Abstract: A device and method are provided that adjust voices of a plurality of sound sources included in distribution content from an information processing device and make it easier to hear sound of each sound source by a reception terminal that receives and reproduces distribution content. A first output voice adjustment unit executes adjustment processing on an output voice of each of the plurality of sound sources, and content including synthesized voice data obtained by synthesizing the output voices corresponding to the sound sources adjusted by the first output voice adjustment unit is output. The first output voice adjustment unit executes output voice adjustment processing for matching a maximum value of a volume level corresponding to a frequency of the output voice of each sound source to a target level. Moreover, a second output voice adjustment unit executes the output voice adjustment processing in accordance with a content type or scene.

Abstract: A computer implemented method for determining if a questioned text has been produced by a human or by a generative artificial intelligence based conversational agent. The method includes the steps of: retrieving from the test text a feature (y) representing the redundancy of the test text; using this feature for determining if the test text has been produced by a human or by a generative artificial intelligence based conversational agent.

Abstract: Some disclosed methods involve: receiving multi-channel audio data including unlabeled multi-channel audio data; extracting audio feature data from the unlabeled multi-channel audio data; applying a spatial masking process to a portion of the audio feature data; applying a contextual encoding process to the masked audio feature data, to produce predicted spatial embeddings in a latent space; obtaining reference spatial embeddings in the latent space; determining a loss function gradient based, at least in part, on a variance between the predicted spatial embeddings and the reference spatial embeddings; and updating the contextual encoding process according to the loss function gradient until one or more convergence metrics are attained.

Abstract: A computer implemented method for the automated analysis or use of data is implemented by a voice assistant. The method comprises the steps of: (a) storing in a memory a structured, machine-readable representation of data that conforms to a machine-readable language (‘machine representation’); the machine representation including representations of user speech or text input to a human/machine interface; and (b) automatically processing the machine representations to analyse the user speech or text input.

Abstract: Provided is a method of generating keyword information by an electronic device. The method includes identifying a text set including at least one text element. The method includes using a named entity recognition (NER) model based on deep learning, identifying keywords in the at least one text element. The method includes based on the text set, determining degrees of association of keyword pairs included in a keyword set including the keywords. The method includes obtaining information on a query word that is input by a user. The method includes based on at least one among the degrees of association, generating information on response words corresponding to the query word.

Abstract: Systems and methods provide for a machine learning system to train a machine learning model to output a penalty-free emission when processing an auditory input. For example, as the system generates paths through a probability lattice, one or more paths may include a penalty-free emission that skips at least one frame associated with the probability lattice, but that does not add a cost to a final path cost. The use of the penalty-free emissions may be represented through one or more graphical representations used for training in order to develop loss functions for models. One or more of these frameworks may be incorporated into automatic speech recognition pipelines to improve training while also reducing coding requirements to simplify debugging operations.

Abstract: The present disclosure includes a system with a first device, a speaker vibrating diaphragm canvas artwork, and a second device featuring a first module and music content. It may also involve a second module, an amplifier, and a smoke detector, the method consists of receiving speech input from an input device in first mode, connected to the output device's first port, requesting artwork content, the input device determines that an artwork response should be displayed on a output device, it then plays an audio message through the diaphragm canvas speaker, indicating that the artwork response will be presented, subsequently, the input device switches to a second mode by sending a control signal to change the active input port to the first port and then sends the artwork response to the output device.

Abstract: A wireless audio device includes a memory including instructions; and a processor operatively connected to the memory and configured to execute the instructions to: detect an audio signal, determine, based on an analysis result of the audio signal, an operation mode of the wireless audio device to be one of a singing mode and a dialogue mode, and control an output signal of the wireless audio device according to the determined operation mode, wherein the dialogue mode is configured to output one or more ambient sounds included in the audio signal, and wherein the singing mode is configured to output one or more media sounds and the one or more ambient sounds included in the audio signal.

Abstract: A method that includes receiving an audio component associated with an audio scene, the audio component including an audio signal, determining a loudness level of the audio component based on the audio signal, receiving a target loudness level for the audio component, producing a bitstream with the audio component by encoding the audio signal and including metadata that has the loudness level and the target loudness level, and transmitting the bitstream to an electronic device.

Abstract: Systems and methods for generative language model tuning can include training the generative language model to generate sets of output text tokens with set of intermediary text tokens with training examples that include input and output pairs. The training can include processing the input with the language model to determine a predicted output and a predicted set of intermediary text tokens. The predicted set of intermediary text tokens can then be evaluated based at least in part on the output associated with the input and the predicted output.

Abstract: In some aspects, the techniques described herein relate to a method including: providing, to a parallel model training platform, a plurality of domain datasets; training, by the parallel model training platform, a plurality of generalist models in parallel, wherein each generalist model of the plurality of generalist models is trained in parallel using a corresponding one of the plurality of domain datasets, and wherein training the plurality of generalist models in parallel generates a corresponding expert model for each generalist model in the plurality of generalist models; executing, by the parallel model training platform, a model parameter averaging process, wherein the model parameter averaging process take each corresponding expert model as input; and generating, by the parallel model training platform and as output of the model parameter averaging process, an average-of-domain-experts (AoDE) model.

Abstract: A computer-implemented method, according to one approach, includes: determining whether initial utterances received from a user include predetermined wake-up words. In response to determining that the initial utterances include the predetermined wake-up words, a voice assistant initiates an interaction with the user. In response to determining that the initial utterances in the audio signal stop for a predetermined period following the initial utterances that include the wake-up words, determining whether additional utterances are received from the user. In response to receiving additional utterances from the user, determining whether the additional utterances have an agitated tone. In situations where the additional utterances do not have an agitated tone, the additional utterances are processed. Furthermore, in response to determining that causing the additional utterances to be processed results in action being taken, the initial utterances are labeled as genuine dialog initiation entries.

Abstract: A system and method for generating a list of tasks from an interaction recording may include a computing device; a memory; and a processor, the processor configured to: identify, for each sentence of one or more sentences of an interaction recording, whether the sentence comprises at least two nouns and at least one verb; generate a set of actionable items using a long short-term memory (LSTM) model by: when the sentence does not comprise at least two nouns and at least one verb, deleting the sentence; and when the sentence comprises at least two nouns and at least one verb, generating an actionable item; input the actionable items to a language model to generate a list of tasks. Computer systems may use a generated list of tasks to automatically allocate work to agents of a contact center or to automatically update a status of a completed task.

Abstract: Techniques for training and testing machine learning models using biased samples are described. In some examples, a sampled test set is generated by estimating a propensity score for each annotation of the set of annotations, wherein a propensity score quantifies a likelihood of being human generated using the set of data, estimating a confidence score for each annotation of the set of annotations, wherein a confidence score quantifies a confidence in a correctness of the annotation, mapping each annotation of the set of annotations to a multi-dimensional space based at least in part on the propensity score, stratifying, based on the propensity score, the mapped annotations, and sampling each stratum according to a request to generate a sampled test set.

Abstract: A method for personalizing responses includes receiving a natural language query from a user that solicits a response from an assistant large language model (LLM) and processing the natural language query to identify, from a log of refinement queries each previously input by the user for instructing the assistant LLM to refine a corresponding previous response generated by the assistant LLM, a preferred refinement query that is relevant to the natural language query. The method also includes prompting, using the natural language query and the preferred refinement query identified as being relevant to the natural language query, the assistant LLM to generate a personalized response to the natural language query. The method also includes providing, for output from a user device, the personalized response to the natural language query.

Abstract: Implementations are directed to efficient federated learning of machine learning (ML) model(s) through on-the-fly decompression and compression of model parameters, of the ML model(s), when facilitating forward propagation and/or back propagation at client device(s). For example, implementations can transmit, from a remote system to a client device, a compressed on-device ML model that includes some compressed parameters. Further, the client device can, in performing forward propagation and/or back propagation using the on-device ML model, decompress those compressed parameters on-the-fly as the parameters are needed for the propagation. The propagation will utilize the decompressed parameters that were decompressed on the fly.