Hyunbae Jeon

Hyunbae Jeon, Jinho D. Choi

As spoken dialogue systems expand beyond traditional assistant roles to encompass diverse personas -- such as authoritative instructors, uncooperative merchants, or distracted workers -- they require distinct, human-like turn-taking behaviors to maintain psychological immersion. However, current full-duplex systems often default to a rigid, overly accommodating ``always-yield'' policy during overlapping speech, which severely undermines character consistency for non-submissive roles. Evaluating alternative, persona-specific turn-taking strategies through empirical user studies is challenging because building real-time full-duplex test environments requires substantial engineering overhead. To address this, we present PersonaKit (PK), an open-source, low-latency web platform for the rapid prototyping and evaluation of conversational agents. Using intuitive JSON configurations, researchers can define personas, specify probabilistic interruption-handling behaviors (e.g., yield, hold, bridge, or override), and automatically deploy comparative A/B surveys. Through an in-the-wild evaluation with 8 distinct personas, we demonstrate that PersonaKit provides an extensible, end-to-end framework for studying complex sociolinguistic behaviors in next-generation spoken agents.

Hyunbae Jeon, Harry He, Anthony Wang et al.

This study investigates game-based learning in the context of the educational game "Jo Wilder and the Capitol Case," focusing on predicting student performance using various machine learning models, including K-Nearest Neighbors (KNN), Multi-Layer Perceptron (MLP), and Random Forest. The research aims to identify the features most predictive of student performance and correct question answering. By leveraging gameplay data, we establish complete benchmarks for these models and explore the importance of applying proper data aggregation methods. By compressing all numeric data to min/max/mean/sum and categorical data to first, last, count, and nunique, we reduced the size of the original training data from 4.6 GB to 48 MB of preprocessed training data, maintaining high F1 scores and accuracy. Our findings suggest that proper preprocessing techniques can be vital in enhancing the performance of non-deep-learning-based models. The MLP model outperformed the current state-of-the-art French Touch model, achieving an F-1 score of 0.83 and an accuracy of 0.74, suggesting its suitability for this dataset. Future research should explore using larger datasets, other preprocessing techniques, more advanced deep learning techniques, and real-world applications to provide personalized learning recommendations to students based on their predicted performance. This paper contributes to the understanding of game-based learning and provides insights into optimizing educational game experiences for improved student outcomes and skill development.

Max Bagga, Hyunbae Jeon, Alex Issokson

Electrocardiography (ECG) signal generation has been heavily explored using generative adversarial networks (GAN) because the implementation of 12-lead ECGs is not always feasible. The GAN models have achieved remarkable results in reproducing ECG signals but are only designed for multiple lead inputs and the features the GAN model preserves have not been identified-limiting the generated signals use in cardiovascular disease (CVD)-predictive models. This paper presents ECGNet which is a procedure that generates a complete set of 12-lead ECG signals from any single lead input using a GAN framework with a bidirectional long short-term memory (LSTM) generator and a convolutional neural network (CNN) discriminator. Cross and auto-correlation analysis performed on the generated signals identifies features conserved during the signal generation-i.e., features that can characterize the unique-nature of each signal and thus likely indicators of CVD. Finally, by using ECG signals annotated with the CVD-indicative features detailed by the correlation analysis as inputs for a CVD-onset-predictive CNN model, we overcome challenges preventing the prediction of multiple-CVD targets. Our models are experimented on 15s 12-lead ECG dataset recorded using MyoVista's wavECG. Functional outcome data for each patient is recorded and used in the CVD-predictive model. Our best GAN model achieves state-of-the-art accuracy with Frechet Distance (FD) scores of 4.73, 4.89, 5.18, 4.77, 4.71, and 5.55 on the V1-V6 pre-cordial leads respectively and shows strength in preserving the P-Q segments and R-peaks in the generated signals. To the best of our knowledge, ECGNet is the first to predict all of the remaining eleven leads from the input of any single lead.

Hyunbae Jeon, Rhea Ramachandran, Victoria Ploerer et al.

Social interactions and conversation skills separate the successful from the rest and the confident from the shy. For college students in particular, the ability to converse can be an outlet for the stress and anxiety experienced on a daily basis along with a foundation for all-important career skills. In light of this, we designed SpeakEasy: a chatbot with some degree of intelligence that provides feedback to the user on their ability to engage in free-form conversations with the chatbot. SpeakEasy attempts to help college students improve their communication skills by engaging in a seven-minute spoken conversation with the user, analyzing the user's responses with metrics designed based on previous psychology and linguistics research, and providing feedback to the user on how they can improve their conversational ability. To simulate natural conversation, SpeakEasy converses with the user on a wide assortment of topics that two people meeting for the first time might discuss: travel, sports, and entertainment. Unlike most other chatbots with the goal of improving conversation skills, SpeakEasy actually records the user speaking, transcribes the audio into tokens, and uses macros-e.g., sequences that calculate the pace of speech, determine if the user has an over-reliance on certain words, and identifies awkward transitions-to evaluate the quality of the conversation. Based on the evaluation, SpeakEasy provides elaborate feedback on how the user can improve their conversations. In turn, SpeakEasy updates its algorithms based on a series of questions that the user responds to regarding SpeakEasy's performance.

Hyunbae Jeon, Frederic Guintu, Rayvant Sahni

Turn-taking prediction is the task of anticipating when the speaker in a conversation will yield their turn to another speaker to begin speaking. This project expands on existing strategies for turn-taking prediction by employing a multi-modal ensemble approach that integrates large language models (LLMs) and voice activity projection (VAP) models. By combining the linguistic capabilities of LLMs with the temporal precision of VAP models, we aim to improve the accuracy and efficiency of identifying TRPs in both scripted and unscripted conversational scenarios. Our methods are evaluated on the In-Conversation Corpus (ICC) and Coached Conversational Preference Elicitation (CCPE) datasets, highlighting the strengths and limitations of current models while proposing a potentially more robust framework for enhanced prediction.