Abstract: Disclosed herein are an apparatus and method for recognizing a voice speaker. The apparatus for recognizing a voice speaker includes a voice feature extraction unit configured to extract a feature vector from a voice signal inputted through a microphone; and a speaker recognition unit configured to calculate a speaker recognition score by selecting a reverberant environment from multiple reverberant environment learning data sets based on the feature vector extracted by the voice feature extraction unit and to recognize a speaker by assigning a weight depending on the selected reverberant environment to the speaker recognition score.

Abstract: Disclosed is a system for language-independent contextual embedding of entities in a document that includes sentences. The system has a database and a processing arrangement. The processing arrangement has a tokenizer module for tokenizing sentences to obtain tokens, an encoder module for determining character coordinate corresponding to the tokens, wherein the character coordinates corresponding to the tokens occur in a multi-dimensional hierarchical space. The system has a transmutation module for processing the character coordinates to generate contextual embeddings thereof in the multi-dimensional hierarchical space and a prediction module for memorizing sequential information pertaining to the contextual embeddings of the character coordinates.

Abstract: Disclosed is a method for delaying audio from a received signal prior to processing in an acoustic echo cancellation device, comprising: receiving an audio signal at an audio extraction and delay device; determining a first amount of delay to apply to the received audio signal such that acoustic echo cancellation (AEC) processing can occur; applying the first amount of delay to the received audio signal; transmitting the un-delayed audio signal to an AEC circuit; and transmitting the delayed audio signal to one or more speakers.

Abstract: Embodiments of the present invention are directed to a computer-implemented method for generating a configurable conversational agent. A non-limiting example of the computer-implemented method includes providing available conversational agents to a user's computing device. The user uses a graphical user interface to choose desired characteristics of a conversational agent. The conversational agent generating service receives the chosen characteristics to generate a customized conversational agent and integrate the conversational agent into the user's system.

Abstract: Disclosed is a microphone system, comprising: one or more microphones each of which is adapted to receive acoustic audio signals and convert the same from acoustical energy signals into electrical audio signals and output the same as microphone output audio signals; and an acoustic echo cancellation (AEC) device adapted to receive the microphone output audio signals, and wherein the AEC device also receives as a second input a reference audio signal from a network, and wherein the AEC device is further adapted to cancel substantially all of the reference audio signal from the microphone output audio signals and output the same as a corrected audio signal, and wherein the reference audio signal comprises an audio signal generated by an external audio system.

Abstract: A method of speech recognition and person identification based thereon, comprising: recording speech from a speech signal using a microphone; illuminating a speaking mouth; recording a degree of light reflected by the mouth from a reflection signal using a sensor; and recording combined parameters of the speech signal and of the reflection signal, and coupling them to letters associated therewith, per predetermined time duration; comparing a combination occurring in speech of parameters of the speech signal and of the reflection signal to the recorded combined parameters of the speech signal and of the reflection signal which are coupled to letters; and deciding on the basis of the comparison to which letter the combination occurring in the speech of parameters of the speech signal and of the reflection signal corresponds, using block-width modulation of the reflection signal.

Abstract: An electronic apparatus and a control method of the electronic apparatus are provided. The electronic apparatus includes: a receiver configured to receive a voice command of a user; and a processor configured to determine a switching probability from a first state corresponding to a current point of time over to a second state corresponding to the voice command, based on the switching probability previously set between a plurality of states related to the electronic apparatus, and selectively perform an operation corresponding to the voice command in accordance with the determined switching probability.

Abstract: Disclosed is a method for determining one or more spoken words, comprising: receiving acoustic audio signals at one or more microphones within a microphone system, and converting the same from acoustical energy signals into electrical audio signals and outputting them as microphone output audio signals; receiving the microphone output audio signals from the microphone device at a first input of an acoustic echo cancellation (AEC) device, and receiving a reference input signal at a second input of the AEC device; cancelling substantially all of the reference audio signal from the microphone output audio signal; and outputting the same as a corrected audio signal, and wherein the reference audio signal comprises an audio signal generated by an external audio system.

Abstract: Disclosed is an audio extraction device for use in an audio signal processing system, comprising: an audio transceiver adapted to receive an audio signal; a delay circuit adapted to delay the received audio signal by a first delay period; a first audio transmitter adapted to transmit the delayed version of the received audio signal to a first destination; and a second audio transmitter adapted to transmit the un-delayed version of the received audio signal to a second destination.

Abstract: Disclosed herein are system, method, and computer program product embodiments for developing natural language skills automatically. In order to generate a skill, metadata is read from a connected service, wherein the metadata specifies entities provided by the service. A relevant entity of the entities provided by the service is determined, and a likely request on the entity is generated.

Abstract: A method to transcribe communications may include obtaining, during a communication session, audio that includes a voice of a user participating in the communication session. The communication session may be configured for verbal communication. The method may further include establishing a network connection with a transcription system and sending the audio to the transcription system. In some embodiments, the transcription system may be configured to generate a transcript of the audio. The method may also include obtaining the transcript of the audio from the transcription system during the communication session and monitoring the audio to determine when the voice is inactive. In some embodiments, in response to the voice being inactive, the method may include stopping the sending of the audio to the transcription system while maintaining the communication session.

Abstract: The innovation disclosed and claimed herein, in one aspect thereof, comprises systems and methods of supporting a conversation with a user. The systems and methods of the innovation can include a bot consumer that receives a first dialogue input from a user, the first dialogue input requesting a response and having an intent. A super-agent selects a worker thread from a pool of worker threads, wherein the worker thread is delegated a task to perform on the first dialogue input. A registrar determines a conversational agent to respond to the user based on the intent. The conversational agent retrieves data to facilitate a response to the first dialogue input and renders a response to the user based on the retrieved data. A registrar custodian can perform create, read, update, and delete operations on the information in the registrar.

Abstract: The invention relates to a method. The method includes receiving a flawed input comprising a domain specific misspelling. The method further includes encoding, by an encoder machine learning model executing on a computer processor, the flawed input on a per character basis to create a context vector. The method further includes decoding, by a decoder machine learning model executing on the computer processor, the context vector on the per character basis to create a rephrased input lacking the domain specific misspelling. The method further includes presenting the rephrased input.

Abstract: A method, computer program product, and system includes a processor(s) obtaining a hashtag and analyzing the hashtag to identify components comprising the hashtag. The processor(s) generate sub-hashtags, where each of the sub-hashtags includes at least one component. The processor(s) obtain, over an Internet connection, via a user interface of a social media platform displayed on a first computing resource, an entry by a user comprising text and the hashtag. The processor(s) obtains from a second computing resource, a request from a second user to access the entry that includes parameters of the second user on the platform. The processor(s) determine that a sub-hashtag is most relevant to the second user. The processor(s) configure instructions to display a customized entry, where the customized entry includes the text and the relevant sub-hashtag for the second user. The processor(s) instruct the second computing resource to display the customized entry in its user interface.

Abstract: Classifying structural features of a digital document by feature type using machine learning is leveraged in a digital medium environment. A document analysis system is leveraged to extract structural features from digital documents, and to classifying the structural features by respective feature types. To do this, the document analysis system employs a character analysis model and a classification model. The character analysis model takes text content from a digital document and generates text vectors that represent the text content. A vector sequence is generated based on the text vectors and position information for structural features of the digital document, and the classification model processes the vector sequence to classify the structural features into different feature types. The document analysis system can generate a modifiable version of the digital document that enables its structural features to be modified based on their respective feature types.

Abstract: One embodiment provides a method, including: obtaining a plurality of chatlogs involving a conversational agent, wherein the plurality of chatlogs represent interactions between users and the conversational agent; identifying dialog states of the conversational agent, wherein each dialog state corresponds to a possible response by the conversational agent; generating a dialog tree including one or more dialog flows of the conversational agent, wherein a dialog flow represents an interaction that occurred between the conversational agent and a user; and providing a conversational reflow user interface for a conversational reflow system allowing execution of the conversational agent to mimic interactions between users and the conversational agent, wherein the conversational reflow system advances through dialog flows based upon input provided by a user at the conversational reflow user interface and the conversational reflow user interface displays output based upon the dialog states of the dialog tree.

Abstract: An apparatus, method, and computer program product for adapting an acoustic model to a specific environment are defined. An adapted model obtained by adapting an original model to the specific environment using adaptation data, the original model being trained using training data and being used to calculate probabilities of context-dependent phones given an acoustic feature. Adapted probabilities obtained by adapting original probabilities using the training data and the adaptation data, the original probabilities being trained using the training data and being prior probabilities of context-dependent phones. An adapted acoustic model obtained from the adapted model and the adapted probabilities.

Abstract: A system, method and computer product for estimating a component of a provided audio signal. The method comprises converting the provided audio signal to an image, processing the image with a neural network trained to estimate one of vocal content and instrumental content, and storing a spectral mask output from the neural network as a result of the image being processed by the neural network. The neural network is a U-Net. The method also comprises providing the spectral mask to a client media playback device, which applies the spectral mask to a spectrogram of the provided audio signal, to provide a masked spectrogram. The media playback device also transforms the masked spectrogram to an audio signal, and plays back that audio signal via an output user interface.

Abstract: Techniques for documenting a clinical procedure involve transcribing audio data comprising audio of one or more clinical personnel speaking while performing the clinical procedure. Examples of applicable clinical procedures include sterile procedures such as surgical procedures, as well as non-sterile procedures such as those conventionally involving a core code reporter. The transcribed audio data may be analyzed to identify relevant information for documenting the clinical procedure, and a text report including the relevant information documenting the clinical procedure may be automatically generated.

Abstract: An engagement classifier for a group chatbot is trained by leveraging the implicit dataset generated by humans engaging in both direct messages as well as group conversations. Human-to-human direct messages are used as an approximate representation of the domain knowledge and expertise of each user. The decision to engage in a group conversation is assumed to be based on that domain knowledge. The knowledge representations and instances of engagements in group conversations yields an effective set of features and labels which can be used to model the engagement decision. The same transfer learning technique is used to generate a knowledge representation for the group chatbot. Given this representation of the domain knowledge of the chatbot, the classifier can predict whether it should engage in any particular group conversation.