Abstract: This application provides a method for training a semantic understanding model, including: obtaining a first training sample set; performing denoising processing on the first training sample set to form a corresponding second training sample set; processing the second training sample set by using a semantic understanding model, to determine initial parameters of the semantic understanding model; processing the second training sample set by using the semantic understanding model in response to the initial parameters of the semantic understanding model, to determine update parameters of the semantic understanding model; and iteratively updating a semantic representation layer network parameter and a task-related output layer network parameter of the semantic understanding model by using the second training sample set and according to the update parameters of the semantic understanding model.

Abstract: Systems and methods for observation-based training of an Artificial Intelligence (AI) character model are provided. An example method includes receiving log data including interactions of a user and a first AI character model, receiving internal parameters of a second AI character model including a first plurality of heuristic machine learning models and a second plurality of primary machine learning models, pre-processing the log data to obtain one or more data streams including behavioral characteristics of the user, running the one or more data streams through the first plurality of heuristic machine learning models to produce intermediate outputs, composing the intermediate outputs into templated formats, and providing the templated formats to the second plurality of primary machine learning models. The internal parameters of the second AI character model are adjusted based on the templated formats such that the second AI character model mimics the behavioral characteristics of the user.

Abstract: Methods, apparatus and articles of manufacture to identify sources of network streaming services are disclosed. An example method includes receiving a first audio signal that represents a decompressed second audio signal, identifying, from the first audio signal, a parameter of an audio compression configuration used to form the decompressed second audio signal, and identifying a source of the decompressed second audio signal based on the identified audio compression configuration.

Abstract: The systems and methods described herein can generate a voice-based interface to increase the accuracy of translations. The voice-based interface can result in fewer input audio signals being transmitted between devices of a network. A method includes: receiving a first input audio signal; generating, based on the first input audio signal, a first translation string in a second language and a second translation string in a first language; determining a first translation score based on a likelihood that the first input audio signal includes an utterance in the first language and a second translation score based on a likelihood that the first input audio signal includes an utterance in the second language; selecting the first translation string based on the first translation score and the second translation score; generating an output signal from the first translation string; and transmitting the output signal to the client device.

Abstract: A speech recognition method and apparatus are disclosed. The speech recognition method includes determining a first score of candidate texts based on an input speech, determining a weight for an output of a language model based on the input speech, applying the weight to a second score of the candidate texts output from the language model to obtain a weighted second score, selecting a target candidate text from among the candidate texts based on the first score and the weighted second score corresponding to the target candidate text, and determining the target candidate text to correspond to a portion of the input speech.

Abstract: A system, method, and computer-program product includes constructing a transcript correction training data corpus that includes a plurality of labeled audio transcription training data samples, wherein each of the plurality of labeled audio transcription training data samples includes: an incorrect audio transcription of a target piece of audio data; a correct audio transcription of the target piece of audio data; and a transcript correction identifier that, when applied to a model input that includes a likely incorrect audio transcript, defines a text-to-text transformation objective causing an audio transcript correction machine learning model to predict a corrected audio transcript based on the likely incorrect audio transcript; configuring the audio transcript correction machine learning model based on a training of a machine learning text-to-text transformer model using the transcript correction training data corpus; and executing the audio transcript correction machine learning model within a speech-to-

Abstract: A computer based system and method for automatically detecting frustration in an interaction, may include: identifying in the interaction using a set of linguistic rules, natural language patterns related to frustration, wherein the linguistic rules further define weights associated with the natural language patterns and rule metadata; reviewing the rule metadata associated with the identified natural language patterns to identify override attributes, wherein if the rule metadata does not include override attributes, then a frustration level in the interaction is determined based on the identified natural language patterns and weights associated with the identified natural language patterns; and if the rule metadata includes override attributes than the frustration level is determined based on the identified override attributes.

Abstract: Techniques for partially processing an input on a device and completing processing at a remote system are provided. The device may process an input using an on-device machine learning (ML) model, and determine to cease processing at an intermediary node of the (ML) model based on the output of the intermediary node. Based on the output of the intermediary node satisfying a condition, the device may use the output of the intermediary node to generate an output responsive to the input. Conversely, if the output of the intermediary node does not satisfy a condition, the device may send the output of the intermediary node to the remote system, so the remote system can use another machine learning model to complete processing with respect to the input.

Abstract: In an illustrative embodiment, systems and methods for automating recorded candidate assessments include receiving a submission for an available position including a question response recording for each of one or more interview questions. For each question response recording, a transcript can be generated by applying a speech-to-text algorithm to an audio portion of the recording. The systems and methods can detect, within the transcript, identifiers each associated with the personality aspects by applying a natural language classifier trained to detect words and phrases associated with the personality aspects of the personality model. Scores may be calculated for each of the personality aspects based on a relevance of the respective personality aspect to the respective interview question and detected identifiers. The scores can be presented within a user interface screen responsive to receiving a request to view interview results.

Abstract: Systems and methods for observation-based training of an Artificial Intelligence (AI) character model are provided. An example method includes receiving log data including interactions of a first user and a second user and adjusting, based on the log data, parameters of the AI character model to cause the AI character model to mimic behavioral characteristics of the first user in follow-up conversations with further users.

Abstract: There is provided methods and apparatuses for decoding and encoding of audio signals. In particular, a method for decoding includes receiving a waveform-coded signal having a spectral content corresponding to a subset of the frequency range above a cross-over frequency. The waveform-coded signal is interleaved with a parametric high frequency reconstruction of the audio signal above the cross-over frequency. In this way an improved reconstruction of the high frequency bands of the audio signal is achieved.

Abstract: An apparatus includes a processor to: receive, from a requesting device, a request to perform speech-to-text conversion of a speech data set; within a first thread of a thread pool, perform a first pause detection technique to identify a first set of likely sentence pauses; within a second thread of the thread pool, perform a second pause detection technique to identify a second set of likely sentence pauses; perform a speaker diarization technique to identify a set of likely speaker changes; divide the speech data set into data segments representing speech segments based on a combination of at least the first set of likely sentence pauses, the second set of likely sentence pauses, and the set of likely speaker changes; use at least an acoustic model with each data segment to identify likely speech sounds; and generate a transcript based, at least in part, on the identified likely speech sounds.

Abstract: A frequency-domain audio codec is provided with the ability to additionally support a certain transform length in a backward-compatible manner, by the following: the frequency-domain coefficients of a respective frame are transmitted in an interleaved manner irrespective of the signalization signaling for the frames as to which transform length actually applies, and additionally the frequency-domain coefficient extraction and the scale factor extraction operate independent from the signalization. By this measure, old-fashioned frequency-domain audio coders/decoders, insensitive for the signalization, would be able to nevertheless operate without faults and with reproducing a reasonable quality. Concurrently, frequency-domain audio coders/decoders able to support the additional transform length would offer even better quality despite the backward compatibility.

Abstract: Disclosed herein are embodiments of systems, methods, and products comprises an analytic server, which automatically manages appointment scheduling. The analytic server receives a customer request to schedule an appointment. The analytic server determines the required data from both customer and service provider for making the appointment. The analytic server retrieves customer data comprising requested service attributes, user preferences, users attributes from internal database and external data source. The analytic server retrieves service providers' data comprising provider service attributes, providers' attributes from internal database and external data sources. The analytic server accesses external data source by web crawling various websites. The analytic server executes an artificial intelligence model to predict user preferences and needs. The analytic server determines potential service providers best matching the customer's input or predicted preferences.

Abstract: An electronic audio file is received that comprises spontaneous speech responsive to a prompt in a non-native language of a speaker. Thereafter, the electronic audio file is parsed into a plurality of spoken words. The spoken words are then normalized to remove stop words and disfluencies. At least one trained content scoring model is then used to determine an absence of pre-defined key points associated with the prompt in the normalized spoken words. A list of the determined absent key points can be generated. This list can then be displayed/caused to be displayed in a graphical user interface along with feedback to improve content completeness. Related apparatus, systems, techniques and articles are also described.

Abstract: A method of processing an acoustic signal is disclosed. According to one or more embodiments, a first acoustic signal is received via a first microphone. The first acoustic signal is associated with a first speech of a user of a wearable headgear unit. A first sensor input is received via a sensor, a control parameter is determined based on the sensor input. The control parameter is applied to one or more of the first acoustic signal, the wearable headgear unit, and the first microphone. Determining the control parameter comprises determining, based on the first sensor input, a relationship between the first speech and the first acoustic signal.

Abstract: An information processing device according to embodiments includes a communication unit configured to receive audio data of content and text data corresponding to the audio data, an audio data reproduction unit configured to perform reproduction of the audio data, a text data reproduction unit configured to perform the reproduction by audio synthesis of the text data, and a controller that controls the reproduction of the audio data or the text data. The controller causes the text data reproduction unit to perform the reproduction of the text data when the audio data reproduction unit is unable to perform the reproduction of the audio data.

Abstract: The situation of a conversation can be allowed to be grasped in more detail. A statement of each participant participating in a conversation is detected by processing a voice signal. The state of each participant participating in the conversation, for example, a direction in which each participant is looking is detected by processing an image signal. The state and existence of a conversation are determined on the basis of the statement of each participant and the state of each participant. The state and existence of a conversation can be determined with higher accuracy than in a technology that determines the state and existence of a conversation only by statements of participants.

Abstract: Automatic measurement of semantic textual similarity of conversations, by: receiving two conversation texts, each comprising a sequence of utterances; encoding each of the sequences of utterances into a corresponding sequence of semantic representations; computing a minimal edit distance between the sequences of semantic representations; and, based on the computation of the minimal edit distance, performing at least one of: quantifying a semantic similarity between the two conversation texts, and outputting an alignment of the two sequences of utterances with each other.

Abstract: NLP techniques are disclosed that apply computer technology to sentence data for performing entity referencing. For example, a processor can parse sentence data in a defined window of sentence data into a list of entity terms and a plurality of classifications associated with the listed entity terms. A processor can also a plurality of context saliency scores for a plurality of the listed entity terms based on the classifications associated with the listed entity terms as well as maintain a list of referring terms corresponding to the listed entity terms. For new sentence data that includes a referring term from the referring term list, a processor can (i) select a corresponding entity term on the entity term list based on the context saliency scores for the entity terms, and (ii) infer that the referring term in the new sentence data refers to the selected corresponding entity term.