arXiv subjects

Youngwon Choi

Publications and source records attributed to Youngwon Choi.

10 recordsLinked to original sources

arXiv2026

Do Spoken Language Models Hear Speech as They Read Text? Bridging Structural Gaps Between Speech and Text

Spoken Language Models (SLMs) generate textual responses directly from speech, offering an alternative to cascaded systems. Despite recent advances, existing SLMs still exhibit weaker instruction-following behavior and limited generalization across diverse tasks compared to text-based language models. Our analysis shows that speech and text representations in current SLMs remain weakly aligned despite strong downstream performance, indicating that structural differences between continuous, temporally varying speech and discrete text remain insufficiently addressed. To address this, we propose a simple framework that decouples length mismatch from semantic alignment and encourages closer correspondence between speech and text representations. Experiments across multiple benchmarks demonstrate competitive performance against strong baselines, underscoring the importance of explicitly addressing structural differences between speech and text in SLM training. Our code is publicly available at https://github.com/jaykim9870/Do_SLMs_Hear_Speech_as_They_Read_Text.

cs.CL↗

arXiv2026

Not All Redundant Tokens Are Alike: Analyzing Visual Token Pruning through Token Roles

Vision-language models (VLMs) process an image as a sequence of visual tokens, which creates a substantial computational bottleneck during inference. Recent visual token pruning methods address this issue by removing seemingly redundant tokens, yet it remains unclear how these pruning decisions relate to the functional roles of visual tokens. In this work, we analyze visual token pruning through the lens of token roles identified by EmbedLens. We first show that representative pruning methods exhibit distinct token-role biases, but these biases do not directly correlate with downstream performance. To better understand this behavior, we refine the token-role assignment procedure and evaluate role-protected pruning variants. Our results show that preserving non-alive tokens can sometimes maintain or improve performance, suggesting that tokens with weak direct semantic alignment may still affect model behavior under pruning. Our code is publicly available at https://github.com/jaykim9870/Not_All_Redundant_Tokens_Are_Alike.

cs.CV↗

arXiv2026

CORTIS: Text-Only Adaptation of Spoken Language Models for Task-Oriented Voice Agents

Task-oriented voice agents need to map spoken user requests to structured outputs such as semantic frames, executable actions, and function calls. A common approach is to cascade ASR with a text-based LLM, but transcription errors can propagate to downstream structured output generation, especially under noisy conditions. Spoken language models (SLMs) offer a direct speech-based alternative, yet adapting them to new tasks typically requires paired speech-target annotations. Motivated by this gap, we present CORTIS, a text-only adaptation framework for task-oriented voice agents. CORTIS fine-tunes SLMs using text-form task supervision, enabling speech-based structured output generation at inference time without task-specific speech-target annotations during adaptation. We evaluate CORTIS on two Qwen2.5-Omni backbones and three task-oriented speech datasets, including an in-house product dataset, and compare it with matched ASR-LLM cascades trained with the same text-form task supervision. Results show that CORTIS performs competitively with matched cascades and offers clearer advantages under acoustic degradation, particularly in preserving high-level task semantics. These findings suggest that text-only fine-tuning of SLMs can serve as a practical adaptation strategy for voice agents when paired speech-target data are costly to collect.

cs.HC↗

arXiv2026

Analyzing Error Propagation in Korean Spoken QA with ASR-LLM Cascades

We analyze how automatic speech recognition (ASR) errors propagate through ASR--LLM cascades in Korean spoken question answering (SQA), focusing on downstream semantic failures that conventional ASR metrics cannot fully capture. Our analysis shows that the relative downstream degradation caused by ASR errors is consistent across LLMs with different absolute performance, suggesting that cascade degradation largely tracks ASR-stage information loss. We further identify single-character ASR errors as a particularly salient source of information loss in Korean, where even a minimal transcription difference can change the intended question and degrade downstream QA performance. Finally, an auxiliary comparison shows that a large audio language model outperforms an ASR--LLM cascade with an approximately matched language backbone in noisy Korean SQA, indicating the potential of direct audio input to mitigate transcript-induced information loss.

cs.CL↗

arXiv2026

Mine-JEPA: In-Domain Self-Supervised Learning for Mine-Like Object Classification in Side-Scan Sonar

Side-scan sonar (SSS) mine classification is a challenging maritime vision problem characterized by extreme data scarcity and a large domain gap from natural images. While self-supervised learning (SSL) and general-purpose vision foundation models have shown strong performance in general vision and several specialized domains, their use in SSS remains largely unexplored. We present Mine-JEPA, the first in-domain SSL pipeline for SSS mine classification, using SIGReg, a regularization-based SSL loss, to pretrain on only 1,170 unlabeled sonar images. In the binary mine vs. non-mine setting, Mine-JEPA achieves an F1 score of 0.935, outperforming fine-tuned DINOv3 (0.922), a foundation model pretrained on 1.7B images. For 3-class mine-like object classification, Mine-JEPA reaches 0.820 with synthetic data augmentation, again outperforming fine-tuned DINOv3 (0.810). We further observe that applying in-domain SSL to foundation models degrades performance by 10--13 percentage points, suggesting that stronger pretrained models do not always benefit from additional domain adaptation. In addition, Mine-JEPA with a compact ViT-Tiny backbone achieves competitive performance while using 4x fewer parameters than DINOv3. These results suggest that carefully designed in-domain self-supervised learning is a viable alternative to much larger foundation models in data-scarce maritime sonar imagery.

cs.CV↗

arXiv2026

ZeSTA: Zero-Shot TTS Augmentation with Domain-Conditioned Training for Data-Efficient Personalized Speech Synthesis

We investigate the use of zero-shot text-to-speech (ZS-TTS) as a data augmentation source for low-resource personalized speech synthesis. While synthetic augmentation can provide linguistically rich and phonetically diverse speech, naively mixing large amounts of synthetic speech with limited real recordings often leads to speaker similarity degradation during fine-tuning. To address this issue, we propose ZeSTA, a simple domain-conditioned training framework that distinguishes real and synthetic speech via a lightweight domain embedding, combined with real-data oversampling to stabilize adaptation under extremely limited target data, without modifying the base architecture. Experiments on LibriTTS and an in-house dataset with two ZS-TTS sources demonstrate that our approach improves speaker similarity over naive synthetic augmentation while preserving intelligibility and perceptual quality. Audio samples are available on our web page.

cs.SD↗

arXiv2025

Exploring Fine-Tuning of Large Audio Language Models for Spoken Language Understanding under Limited Speech Data

Large Audio Language Models (LALMs) have emerged as powerful tools for speech-related tasks but remain underexplored for fine-tuning, especially with limited speech data. To bridge this gap, we systematically examine how different fine-tuning schemes including text-only, direct mixing, and curriculum learning affect spoken language understanding (SLU), focusing on scenarios where text-label pairs are abundant while paired speech-label data are limited. Results show that LALMs already achieve competitive performance with text-only fine-tuning, highlighting their strong generalization ability. Adding even small amounts of speech data (2-5%) yields substantial further gains, with curriculum learning particularly effective under scarce data. In cross-lingual SLU, combining source-language speech data with target-language text and minimal target-language speech data enables effective adaptation. Overall, this study provides practical insights into the LALM fine-tuning under realistic data constraints.

cs.SD↗

arXiv2025

DESAMO: A Device for Elder-Friendly Smart Homes Powered by Embedded LLM with Audio Modality

We present DESAMO, an on-device smart home system for elder-friendly use powered by Audio LLM, that supports natural and private interactions. While conventional voice assistants rely on ASR-based pipelines or ASR-LLM cascades, often struggling with the unclear speech common among elderly users and unable to handle non-speech audio, DESAMO leverages an Audio LLM to process raw audio input directly, enabling a robust understanding of user intent and critical events, such as falls or calls for help.

cs.HC↗

arXiv2023

Wasserstein Geodesic Generator for Conditional Distributions

Generating samples given a specific label requires estimating conditional distributions. We derive a tractable upper bound of the Wasserstein distance between conditional distributions to lay the theoretical groundwork to learn conditional distributions. Based on this result, we propose a novel conditional generation algorithm where conditional distributions are fully characterized by a metric space defined by a statistical distance. We employ optimal transport theory to propose the Wasserstein geodesic generator, a new conditional generator that learns the Wasserstein geodesic. The proposed method learns both conditional distributions for observed domains and optimal transport maps between them. The conditional distributions given unobserved intermediate domains are on the Wasserstein geodesic between conditional distributions given two observed domain labels. Experiments on face images with light conditions as domain labels demonstrate the efficacy of the proposed method.

stat.ML↗

arXiv2022

SimpleMind adds thinking to deep neural networks

Deep neural networks (DNNs) detect patterns in data and have shown versatility and strong performance in many computer vision applications. However, DNNs alone are susceptible to obvious mistakes that violate simple, common sense concepts and are limited in their ability to use explicit knowledge to guide their search and decision making. While overall DNN performance metrics may be good, these obvious errors, coupled with a lack of explainability, have prevented widespread adoption for crucial tasks such as medical image analysis. The purpose of this paper is to introduce SimpleMind, an open-source software framework for Cognitive AI focused on medical image understanding. It allows creation of a knowledge base that describes expected characteristics and relationships between image objects in an intuitive human-readable form. The SimpleMind framework brings thinking to DNNs by: (1) providing methods for reasoning with the knowledge base about image content, such as spatial inferencing and conditional reasoning to check DNN outputs; (2) applying process knowledge, in the form of general-purpose software agents, that are chained together to accomplish image preprocessing, DNN prediction, and result post-processing, and (3) performing automatic co-optimization of all knowledge base parameters to adapt agents to specific problems. SimpleMind enables reasoning on multiple detected objects to ensure consistency, providing cross checking between DNN outputs. This machine reasoning improves the reliability and trustworthiness of DNNs through an interpretable model and explainable decisions. Example applications are provided that demonstrate how SimpleMind supports and improves deep neural networks by embedding them within a Cognitive AI framework.

cs.AI↗