Abstract: The present disclosure provides method and apparatus for responding in a voice conversation by an electronic conversational agent. A voice input may be received in an audio upstream. In response to the voice input, a primary response and at least one supplementary response may be generated. A primary voice output may be generated based on the primary response. At least one supplementary voice output may be generated based on the at least one supplementary response. The primary voice output and the at least one supplementary voice output may be provided in an audio downstream, wherein the at least one supplementary voice output is provided during a time period adjacent to the primary voice output in the audio downstream.

Abstract: The present disclosure is directed to systems and methods for performing reading comprehension with machine learning. More specifically, the present disclosure is directed to a Neural Symbolic Reader (example implementations of which may be referred to as NeRd), which includes a reader to encode the passage and question, and a programmer to generate a program for multi-step reasoning. By using operators like span selection, the program can be executed over a natural language text passage to generate an answer to a natural language text question. NeRd is domain-agnostic such that the same neural architecture works for different domains. Further, NeRd is compositional such that complex programs can be generated by compositionally applying the symbolic operators.

Abstract: The present disclosure is directed to a device and method for detecting presence or absence of human speech. The device and method utilize a low-power accelerometer. The device and method generate an acceleration signal using the accelerometer, filter the acceleration signal with a band pass filter or a high pass filter, determine at least one calculation of the filtered acceleration signal, detect a presence or absence of a voice based on the at least one calculation, and output a detection signal that indicates the presence or absence of the voice. The device and method are well suited for portable audio devices, such as true wireless stereo headphones, that have a limited power supply.

Abstract: Generating questions by receiving user utterance data, determining an intent confidence vector for the user utterance data, predicting, by a trained next user-intent prediction model, a next user-intent confidence vector using the intent confidence vector, and generating a next question using the next user-intent confidence vector.

Abstract: A method for training a non-autoregressive TTS model includes receiving training data that includes a reference audio signal and a corresponding input text sequence. The method also includes encoding the reference audio signal into a variational embedding that disentangles the style/prosody information from the reference audio signal and encoding the input text sequence into an encoded text sequence. The method also includes predicting a phoneme duration for each phoneme in the input text sequence and determining a phoneme duration loss based on the predicted phoneme durations and a reference phoneme duration. The method also includes generating one or more predicted mel-frequency spectrogram sequences for the input text sequence and determining a final spectrogram loss based on the predicted mel-frequency spectrogram sequences and a reference mel-frequency spectrogram sequence. The method also includes training the TTS model based on the final spectrogram loss and the corresponding phoneme duration loss.

Abstract: A description of a conversation may be generated to allow a person to understand important aspects of the conversation without needing to review the conversation. The conversation description may be generated by identifying one or more events that occurred in the conversation and then generating the description using the identified events. A set of possible events may be determined in advance for a particular application. The events may be identified by using an event neural network for each event. Each event neural network may process the messages of the conversation to generate an event score that indicates a match between the conversation and the corresponding event. The event scores may then be used to select one or more events. Message scores from the event neural network of a selected event may then be used to select one or more messages of the conversation as a rationale for the selected event.

Abstract: In an approach to improve chatbot workspaces by updating chatbot workspaces through documentation updating and chatbot skill updating. Embodiments determine a chatbot knowledge base contains a set of updated information and updates a chatbot dialog decision tree based on one or more identified new topics in a set of updated information using natural language processing techniques to determine a set of intents, a set of entities, and a set of keywords. Further, embodiments identify a starting decision for traversing the chatbot dialogue decision tree based on the updated set of entities and the updated set of keywords. Additionally, embodiments interact, via a user interface, with an end user according to one or more interactions traversing the chatbot dialogue decision tree for a response.