Abstract: Speech and text analysis processing for detecting computer-generated speech is provided. Speech and audio from a voice call or other interaction between two entities may be monitored and analyzed using one or more machine models. Audio may be transcribed and the resulting words and phrases used in the audio analyzed by a machine model to determine a likelihood that the audio is computer-generated. The audio may be separately analyzed to evaluate characteristics such as tones, inflections, accents, pitch, pace, and the like to determine a further likelihood of whether the audio is computer-generated. A duration of the audio may be used as a scoring factor. The various probabilities and scores may be combined to provide a composite score.

Abstract: Some embodiments provide retail product ordering systems comprising: a user computing device comprising an application executed by a device control circuit to: receive an audible utterance; controls a product identifier application interface to: apply a tokenizer model and obtain a set of individual search words; apply a series of featurizer models to the search words to generate features; and apply a classifier and extractor model based on the features and generate multiple requested product entities each comprising a respective sub-set of the position labeled product terms; wherein the device control circuit is further configured to access a purchase history database, confirm an accuracy of each of requested product entities relative to a purchase history, generate a listing of determined product identifiers corresponding to the confirmed set of the multiple requested product entities, and control a display system of the user computing device to render the listing of determined product identifiers.

Abstract: A speech recognition device is a speech recognition device for performing speech recognition using an end-to-end model and includes an encoder that is a model that converts an input speech signal into a characteristic expression, a decoder that is a model that converts speech data into text using an output of the encoder, and a learning unit that randomly selects a block length of the speech signal input to the encoder and causes the encoder and the decoder to learn.

Abstract: The technology discloses a system for validation of generative AI model operators using preloaded query context connected to independent AI models in generative AI-driven trivia gameplay. The system includes a user interface connected to a generative AI model that generates trivia responses based on user input. A model superstructure between the user interface, the generative AI model, and independent validation model(s) facilitates the evaluation of the user input or trivia response by the validation model(s). The validation model(s) are provided with a pre-loaded query context to detect adherence to predetermined conditions. A consensus module receives the validation model's output and delivers the user input or trivia response to the generative AI model or user interface, respectively, based on the evaluation.

Abstract: Various embodiments of the present disclosure provide machine learning training techniques for implementing a multi-modal interpretation process to generate holistic outputs for an event. The techniques may include generating, using first layers of a multi-modal machine learning model, text-based intermediate representations for an entity based on textual input data. The techniques include generating, using second layers of the multi-modal machine learning model, image-based intermediate representations for the entity based on the text-based intermediate representations and input images for the entity. The techniques include generating, using one or more third layers of the multi-modal machine learning model, an entity representation summary based on the image-based intermediate representations and an image narrative summary for the input images. The techniques include initiating the performance of a prediction-based action based on the entity representation summary.

Abstract: In an example method, a system accesses first data including text transcripts of a plurality of voice calls, and generates, based on the first data, labeled representations of the voice calls using the one or more computerized LLMs. The system generates the labeled representations by determining a plurality of contextual categories associated with the voice calls, segmenting the text transcript into a plurality of transcript segments, and associating each of the transcript segments with a respective one of the contextual categories. Further, the system generates second data representing the labeled representations of the voice calls and stores the second data using the one or more hardware storage devices.

Abstract: A computer-implemented system and computer instructions stored on non-transitory computer readable medium for bilingual medical inquiry in both Arabic and English, including multiple-choice question answering, open-ended question answering, and multi-turn question answering. The system and instructions use a mixture of experts large language model (MOE LLM) having a router network connected to multiple expert networks. The MOE LLM is trained with medical domain data and is used to receive the input bilingual text in a format for a medical inquiry, and output text in a format of a response to the medical inquiry, in sequence. The system and instructions incorporate an English-to-Arabic translation pipeline having a language translation model to generate Arabic language medical instruction sets from English language medical instructions, for large scale use in Arabic and English medical inquiry.

Abstract: A caption of a multimodal message (e.g., social media post) can be identified as a named entity using an entity recognition system. The entity recognition system can use a visual attention based mechanism to generate a visual context representation from an image and caption. The system can use the visual context representation to identify one or more terms of the caption as a named entity.

Abstract: Aspects of the disclosure are directed to a transactional agent for user interactions. The agent can seamlessly respond to user requests in a conversational manner while maintaining the conversational state. The agent can include a multi-stage modular model architecture, including a semantic understander and a semantic matcher. The semantic understander can be configured to understand common conversation conventions and/or patterns to produce a structure representation of a user request. The semantic matcher can be configured to map items and modifiers to product entries for a particular domain.

Abstract: An illustrative intent classification engine may access a text transcript and determine one or more features associated with the text transcript. Based on the one or more features, the intent classification engine may generate an aggregate embedding vector and provide the aggregate embedding vector as an input to a trained model configured to output an intent classification. Corresponding methods and systems are also disclosed.

Abstract: Disclosed herein are devices, systems, and computer-implemented methods for intelligent conversational intent detection. Example methods include acquiring a conversational transcript input that is requested for intent detection, inputting the conversational transcript input into a model configured to decipher a conversational intent segment, and returning to a user the conversational intent segment. The conversational transcript input can include one or more conversational transcript segments. The conversational intent segment can correspond to which of the one or more conversational transcript segments is likely to indicate a conversational intent of the conversational transcript input based on the intent detection.

Abstract: An example operation may include one or more of executing an interaction with an account via a chatbot within a chat element running on an account device, determining a next response for the chatbot to output within the chat element based on an execution of an artificial intelligence (AI) model on content within the chat element, determining an interaction attribute of the next response based on an execution of a second AI model on the content from the chat element and on historical chat content of the account device and the chatbot, and outputting the next response via the chatbot within the chat element based on the determined interaction attribute. The example operation may further include an AI agent that performs an action based on the next response.

Abstract: A text analysis processing for detecting computer-generated text is provided. In some cases, a text-based chat interaction may be initiated and analyzed to determine whether the text-based chat generated by a communicating entity is computer-generated. The text of the chat session may be analyzed to evaluate punctuation, use of emojis, spacing, grammar, words, phrases, and the like to determine a further likelihood of whether the text is computer-generated. A duration of the chat session may be used as a scoring factor. The various probabilities and scores may be combined to provide a composite score.

Abstract: Verbal language analysis is provided to users. The user enrolls or subscribes for verbal language analysis or analytics. The user carries out or conducts a conversation with a third party. An intelligence device associated with the user records the conversation. The intelligence device performs verbal language analysis on the conversation. The verbal language analysis generates individual metrics for verbal factors of energy, word count, inflection, tone (e.g. pitch and sentiment), rate, and/or the like. A verbal intelligence index is determined from the individual metrics using aggregation, averaging, weighted averaging, and/or the like. An interface component generates views to display to the user for review of the conversation to facilitate better verbal performance during current and in future conversations.

Abstract: Systems and methods for recording and transcribing conversations in real-time. Sentiment analysis is performed on each utterance to determine both an intent of the conversation along with sentiment. Annotated transcripts are provided. A machine learning model may be used to augment sentiment analysis.

Abstract: A device includes a processor, and a memory storing executable instructions which, when executed by the processor, cause the processor alone or in combination with other processors to perform the following functions: receive textual user input from a user describing a design to be generated; implement a first prompt generator to generate a first prompt for a Large Language Model (LLM) to restructure the user input; and implement a second prompt generator to generate a second prompt for a text-to-image model using output of the LLM to produce, the second prompt to prompt the text-to-image model to produce a proposed design based on the user input. The proposed design is provided to the user via an application comprising controls for further editing the proposed design.

Abstract: Devices and techniques are described for embedding-free speaker diarization. In some examples, a first speaker ID label is determined for a first frame and a second speaker ID label may be determined for a second frame of a first window of audio. A third speaker ID label may be determined for a third frame of a second window. First combined data representing at least the first frame and the third frame and second combined data representing at least the second frame and the third frame may be generated. First posterior data associated with the first frame and second posterior data associated with the third frame may be generated. Third posterior data associated with the second frame and fourth posterior data associated with the third frame may be generated. A determination may be made that the first speaker ID label and the third speaker ID label correspond to the same speaker.

Abstract: Methods and apparatuses for optimizing performance of conversational interface applications using example forgetting include a server that retrieves training data comprising utterances each mapped to one or more known intents. The server determines a forgetting count for each utterance and selects utterances from the training data that have a forgetting count above a predetermined threshold. The server identifies whether the predicted intent associated with each utterance is accurate. The server generates updated training data comprising the selected utterances and corresponding predicted intents, and trains conversational interface applications using the updated training data. The server validates performance of the trained conversational interface applications and saves the updated training data.

Abstract: The information processing device according to one embodiment includes: a setting part configured to pair and set, based on first information sent from a terminal that is connected via a communication network, user identification information with second information, the user identification information identifying a user using a given telephone machine, the second information being obtainable from a voice packet of a telephone call between the given telephone machine and another telephone machine, and the first information being predetermined information and the second information being predetermined information; and a specifying part configured to obtain, when a voice packet of a telephone call between the given telephone machine and another telephone machine arrives, the second information from the voice packet, and specify the user identification information paired with the second information obtained.

Abstract: A method of leveraging machine learning to predict empathy for improved contact center interactions according to an embodiment includes receiving, by a computing system, at least one user message from a real-time contact center interaction with a user, generating, by an artificial intelligence system of the computing system, at least one empathy score based on the at least one message using the machine learning, wherein each of the at least one empathy score is indicative of a real-time empathy of the user, generating, by the artificial intelligence system of the computing system, an empathetic text response to the at least one user message based on the at least one empathy score, and responding to the at least one user message in the real-time contact center interaction based on the empathetic text response generated by the artificial intelligence system of the computing system.