Abstract: A masking system prevents a human agent from receiving sensitive personal information (SPI) provided by a caller during caller-agent communication. The masking system includes components for detecting the SPI, including automated speech recognition and natural language processing systems. When the caller communicates with the agent, e.g., via a phone call, the masking system processes the incoming caller audio. When the masking system detects SPI in the caller audio stream or when the masking system determines a high likelihood that incoming caller audio will include SPI, the caller audio is masked such that it cannot be heard by the agent. The masking system collects the SPI from the caller audio and sends it to the organization associated with the agent for processing the caller's request or transaction without giving the agent access to caller SPI.

Abstract: A computer-implemented method for training a neural network that is configured to generate a score distribution over a set of multiple output positions. The neural network is configured to process a network input to generate a respective score distribution for each of a plurality of output positions including a respective score for each token in a predetermined set of tokens that includes n-grams of multiple different sizes. Example methods described herein provide trained neural networks which produce results with improved accuracy compared to the state of the art, e.g. translations that are more accurate compared to the state of the art, or more accurate speech recognition compared to the state of the art.

Abstract: Methods, systems, and computer program products are provided for language entity identification. In one embodiment, a computer-implemented method is disclosed. In the method, respective pinyin codes may be determined for respective Chinese characters comprised in a string that is to be processed. Then, respective pinyin features may be generated from the respective pinyin codes. Next, a candidate language entity may be identified from the string based on the respective pinyin features and a mapping describing an association between pinyin features and language entity. In other embodiments, a computer-implemented system and a computer program product for security management are disclosed.

Abstract: The present disclosure discloses method and a user intent identification system for identifying user intent from user statements. The user intent identification system receives input statement provided by a user from a Natural Language Understanding (NLU) engine. The input statement is processed to remove one or more irrelevant content. A plurality of features for each word in the processed input statement is extracted. The plurality of features comprises Parts of Speech (POS) label, dependency parse tree and word embeddings. The user intent determination system predicts class for each word in the processed input statement from a plurality of predefined classes using a neural network model. The neural network model predicts class for each word based on input vector generated for the each word based on the plurality of features. Thereafter, the user intent is identified based on class predicted for each word in processed input statement.

Abstract: Techniques related to keyphrase detection for applications such as wake on voice are disclosed herein. Such techniques may have high accuracy by using scores of phone positions in triphones to select which triphones to use with a rejection model, using context-related phones for the rejection model, adding silence before keyphrase sounds for a keyphrase model, or any combination of these.

Abstract: The present disclosure relates to an image processing device, an operation method of the image processing device, and a computer-readable recording medium. The image processing device according to an embodiment in the present disclosure may comprise: a voice-obtaining unit for obtaining the voice of a user and generating a first voice signal; a communication interface unit for receiving a second voice signal of the user from an external device; and a processor which, after the first voice signal is received from the voice-obtaining unit, performs a first pre-processing operation employing voice amplification of the received first voice signal, and, after the second voice signal is received via the communication interface unit, performs a second pre-processing operation employing noise amplification of the second voice signal.

Abstract: The invention discloses a training method and a speech recognition method for a mixed frequency acoustic recognition model, which belongs to the technical field of speech recognition.

Abstract: A computer-implemented method comprises: receiving a first address and a second address, the first address including n morphemes and the second address including m morphemes, wherein a morpheme is a smallest semantic unit in the address, and wherein n and m are both natural numbers; determining first correlation values between the n morphemes and the m morphemes; obtaining, based on the first correlation values and a preset algorithm, a second correlation value between the first address and the second address; and analyzing a correlation between the first address and the second address based on the second correlation value.

Abstract: Agents are coached to improve their performance by participation in a speech coaching campaign. In one embodiment, an administrator identifies top and bottom performing agents, and retrieves their voice call recordings that are processed by a speech analytics system to produce word clouds corresponding to desirable and undesirable phrases. After reviewing and potentially editing the word clouds, a set of desirable and undesirable operational phrases are created, which the agent should use, or not use, during a call. A speech analytics system is configured to detect the presence of these operational phrases for an agent when the agent is on a call. The agent may review information depicting how well they are utilizing the desirable phrases and avoiding the undesirable phrases, and points may be allocated reflecting the agent's usage. The points may be processed by a gamification system to incentivize the agent to improve their performance.

Abstract: A system for artificial intelligent dispute resolution is disclosed. The system may receive a dispute initiation request from a voice input channel. The system may determine user authentication state in response to the dispute initiation request. The system may receive a natural language problem statement from the voice input channel. The system may determine a user intent in response to the natural language problem statement. The system may compare the user intent with a business rules set and determine a dispositioned outcome based on the business rules set and the user intent.

Abstract: Techniques are provided for dynamic adaptation of language understanding systems to acoustic environments. A methodology implementing the techniques according to an embodiment includes generating a trigger in response to recognition of a wake-on-voice key-phrase in or prior to an audio stream. The trigger serves to switch processing modes from an adaptation mode to a query recognition mode. The method further includes performing automatic speech recognition on the audio stream during the query recognition mode, to recognize an in-domain query. The method further includes applying both a static language understanding classifier and a dynamic language understanding classifier to the recognized in-domain query. The static language understanding classifier employs a static semantic model and the dynamic language understanding classifier employs a dynamic semantic model.

Abstract: A method and system may select an interaction involving an agent, sending the selected interaction and a computerized form to the agent and an evaluator, simultaneously or concurrently, displaying to the evaluator and agent screens defined by the form, each screen including an evaluation question, accepting from the agent, for each evaluation question, an agent answer having associated with the agent answer a rating, accepting from the evaluator a submission indicating that the evaluator has completed the computerized evaluation form, accepting from the agent a submission indicating that the agent has completed the computerized evaluation form, summing an agent rating from the ratings associated with the agent answers provided by the agent, summing an evaluator rating from the ratings associated with evaluator answers provided by the evaluator, and calculating a variance from the agent rating and evaluator rating.

Abstract: The embodiments of the disclosure disclose an intelligent playing method and apparatus based on preference feedback. An embodiment of the method comprises: receiving voice feedback on currently played multimedia from a user; analyzing a user intention based on the voice feedback; calculating, in response to the user intention indicating updating a currently played multimedia list, a similarity between multimedia in a multimedia database and the current multimedia; and updating the currently played multimedia list based on the voice feedback and the similarity. The embodiment improves the quality and pertinence in playing multimedia.

Abstract: Techniques for avoiding wake word self-triggering are provided. In one embodiment, an electronic device can receive an audio-out signal to be output as audio via a speaker of the device and can attempt to recognize a wake word in the audio-out signal using a first recognizer. If the wake word is recognized in the audio-out signal, the electronic device can further determine whether a wake word match is made using a second recognizer with respect to a mic-in audio signal captured via a microphone of the device at approximately the same time that the audio-out signal is output via the speaker. If so, the electronic device can ignore the wake word match made using the second recognizer.

Abstract: A chatbot maintains a conversation with a user by providing a plurality of separate conversational tiers that allow the chatbot to interact with the user, each of the conversational tiers having a set of production rules that are independent of production rules for other ones of the tiers. The production rules indicate a flow of conversation between the user and the chatbot. The chatbot selects one of the conversational tiers based on previous conversational tiers used by the chatbot in connection with conversing with the user and based on content of conversations between the user and the chatbot. The chatbot responds to statements of the user according to a specific production rule that is chosen by the chatbot based on a particular one of the conversational tiers that has been selected and based on other factors. A scripting engine may match statements of the user with specific production rules.

Abstract: A voice activated assistant activation prevention system includes a database storing audio source information describing a relative location of an audio source. The system is configured to monitor, using at least one microphone, for an activation command that is associated with a recording of a subsequent instruction command and a sending that subsequent instruction command through a network. When the system detects, using first audio information received by the at least one microphone, a first instance of the activation command, it determines a source location of the first instance of the activation command. If the system matches the source location of the first instance of the activation command to the relative location of the audio source described by the audio source information in the database, the system may prevent the sending though the network of an instruction command that immediately follows the first instance of the activation command.

Abstract: An automatic speech recognition (ASR) system includes at least one processor and a memory storing instructions.

Abstract: A system and method to receive a spoken utterance and convert the spoken utterance into a recognized speech results through an automatic speech recognition service. A spoken utterance into a recognized speech result through an automatic speech recognition service. The recognized speech results are interpreted through a natural language processing module. A normalizer processes the recognized speech results that transforms the recognized speech interpretations into predefined form for a given automatic speech recognition domain and further determines which automatic speech recognition domains or the recognized speech results are processed by a dedicated dialogue management proxy module or a conversation module.

Abstract: An electronic apparatus is provided. The electronic apparatus includes a memory, a microphone and a processor configured to compare a volume of a voice input through the microphone and a standard voice volume stored in the memory, corresponding to a space in which the electronic apparatus is located, and identify whether to perform a voice recognition on the voice based on the comparison.

Abstract: The present disclosure provides a method, computer system and computer program product facilitating generation of a conversation content. A computer-implemented method comprises: obtaining, by a device operatively coupled to one or more processors, history conversation data of an user; determining, by the device, a professional level of the user based on the history conversation data; and generating, by the device, a conversation content for the user based on the professional level of the user.