arXiv subjects

Hung-yi Lee

Publications and source records attributed to Hung-yi Lee.

At least 91 records · Page 5Linked to original sources

arXiv2026

MoVE: Translating Laughter and Tears via Mixture of Vocalization Experts in Speech-to-Speech Translation

Recent Speech-to-Speech Translation (S2ST) systems achieve strong semantic accuracy yet consistently strip away non-verbal vocalizations (NVs), such as laughter and crying that convey pragmatic intent, which severely limits real-world utility. We address this via three contributions. First, we propose a synthesis pipeline for building scalable expressive datasets to overcome the data scarcity limitation. Second, we propose MoVE, a Mixture-of-LoRA-Experts architecture with expressive-specialized adapters and a soft-weighting router that blends experts for capturing hybrid expressive states. Third, we show pretrained AudioLLMs enable striking data efficiency: 30 minutes of curated data is enough for strong performance. On English-Chinese S2ST, while comparing with strong baselines, MoVE reproduces target NVs in 76% of cases and achieves the highest human-rated naturalness and emotional fidelity among all compared systems, where existing S2ST systems preserve at most 14% of NVs.

cs.CL↗

arXiv2026

Style Amnesia: Investigating Speaking Style Degradation and Mitigation in Multi-Turn Spoken Language Models

In this paper, we show that when spoken language models (SLMs) are instructed to speak in a specific speaking style at the beginning of a multi-turn conversation, they cannot maintain the required speaking styles after several turns of interaction; we refer to this as the style amnesia of SLMs. We focus on paralinguistic speaking styles, including emotion, accent, volume, and speaking speed. We evaluate three proprietary and two open-source SLMs, demonstrating that none of these models can maintain a consistent speaking style when instructed to do so. We further show that while SLMs can recall the style instruction when prompted in later turns, they still fail to express it, but through explicit recall can mitigate style amnesia. In addition, SLMs struggle more when the style instruction is placed in system messages rather than user messages, even though system messages are specifically designed to provide persistent, conversation-level instructions. Our findings highlight a systematic gap in current SLMs' ability to maintain speaking styles, highlighting the need for improved style adherence in future models. Our code and evaluation data are publicly available at https://github.com/YuXiangLin1234/SLM-Style-Amnesia.

cs.CL↗

arXiv2026

ASPIRin: Action Space Projection for Interactivity-Optimized Reinforcement Learning in Full-Duplex Speech Language Models

End-to-end full-duplex Speech Language Models (SLMs) require precise turn-taking for natural interaction. However, optimizing temporal dynamics via standard raw-token reinforcement learning (RL) degrades semantic quality, causing severe generative collapse and repetition. We propose ASPIRin, an interactivity-optimized RL framework that explicitly decouples when to speak from what to say. Using Action Space Projection, ASPIRin maps the text vocabulary into a coarse-grained binary state (active speech vs. inactive silence). By applying Group Relative Policy Optimization (GRPO) with rule-based rewards, it balances user interruption and response latency. Empirical evaluations show ASPIRin optimizes interactivity across turn-taking, backchanneling, and pause handling. Crucially, isolating timing from token selection preserves semantic coherence and reduces the portion of duplicate n-grams by over 50% compared to standard GRPO, effectively eliminating degenerative repetition.

cs.CL↗

arXiv2026

Joint Fullband-Subband Modeling for High-Resolution SingFake Detection

Rapid advances in singing voice synthesis have increased unauthorized imitation risks, creating an urgent need for better Singing Voice Deepfake (SingFake) Detection, also known as SVDD. Unlike speech, singing contains complex pitch, wide dynamic range, and timbral variations. Conventional 16 kHz-sampled detectors prove inadequate, as they discard vital high-frequency information. This study presents the first systematic analysis of high-resolution (44.1 kHz sampling rate) audio for SVDD. We propose a joint fullband-subband modeling framework: the fullband captures global context, while subband-specific experts isolate fine-grained synthesis artifacts unevenly distributed across the spectrum. Experiments on the WildSVDD dataset demonstrate that high-frequency subbands provide essential complementary cues. Our framework significantly outperforms 16 kHz-sampled models, proving that high-resolution audio and strategic subband integration are critical for robust in-the-wild detection.

cs.SD↗

arXiv2026

Full-Duplex-Bench-v3: Benchmarking Tool Use for Full-Duplex Voice Agents Under Real-World Disfluency

We introduce Full-Duplex-Bench-v3 (FDB-v3), a benchmark for evaluating spoken language models under naturalistic speech conditions and multi-step tool use. Unlike prior work, our dataset consists entirely of real human audio annotated for five disfluency categories, paired with scenarios requiring chained API calls across four task domains. We evaluate six model configurations -- GPT-Realtime, Gemini Live 2.5, Gemini Live 3.1, Grok, Ultravox v0.7, and a traditional Cascaded pipeline (Whisper$\rightarrow$GPT-4o$\rightarrow$TTS) -- across accuracy, latency, and turn-taking dimensions. GPT-Realtime leads on Pass@1 (0.600) and interruption avoidance (13.5\%); Gemini Live 3.1 achieves the fastest latency (4.25~s) but the lowest turn-take rate (78.0\%); and the Cascaded baseline, despite a perfect turn-take rate, incurs the highest latency (10.12~s). Across all systems, self-correction handling and multi-step reasoning under hard scenarios remain the most consistent failure modes.

eess.AS↗

arXiv2026

The Binding Effect: Analyzing How Multi-Dimensional Cues Form Gender Bias in Instruction TTS

Current bias evaluations in Instruction Text-to-Speech (ITTS) often rely on univariate testing, overlooking the compositional structure of social cues. In this work, we investigate gender bias by modeling prompts as combinations of Social Status, Career stereotypes, and Persona descriptors. Analyzing open-source ITTS models, we uncover systematic interaction effects where social dimensions modulate one another, creating complex bias patterns missed by univariate baselines. Crucially, our findings indicate that these biases extend beyond surface-level artifacts, demonstrating strong associations with the semantic priors of pre-trained text encoders and the skewed distributions inherent in training data. We further demonstrate that generic diversity prompting is insufficient to override these entrenched patterns, underscoring the need for compositional analysis to diagnose latent risks in generative speech.

eess.SP↗

arXiv2026

DeSTA2.5-Audio: Toward General-Purpose Large Audio Language Model with Self-Generated Cross-Modal Alignment

We introduce DeSTA2.5-Audio, a general-purpose Large Audio Language Model (LALM) designed for robust auditory perception and instruction-following. Recent LALMs augment Large Language Models (LLMs) with auditory capabilities by training on large-scale audio-instruction datasets. However, existing LALMs have often suffered from the catastrophic forgetting of the LLM's original abilities. Therefore, balancing knowledge retention and audio perception has become a critical challenge. To address this, we revisit the data construction pipeline and propose a self-generated cross-modal alignment strategy in which the backbone LLM generates its own training targets, named DeSTA. This approach aims at preserving the LLM's native language proficiency thereby enabling zero-shot generalization without task-specific tuning. We construct DeSTA-AQA5M, a large-scale, task-agnostic dataset containing 5 million training samples derived from 7,000 hours of audio spanning 50 diverse datasets, including speech, environmental sounds, and music. DeSTA2.5-Audio achieves state-of-the-art or competitive performance across a wide range of audio-language benchmarks, including Dynamic-SUPERB, MMAU, SAKURA, Speech-IFEval, and VoiceBench. Comprehensive comparative studies demonstrate that our self-generated strategy outperforms existing training strategies. Our findings underscore the importance of carefully designed data construction in LALM development and offer practical insights for building robust, general-purpose LALMs.

eess.AS↗

arXiv2026

How Auditory Knowledge in LLM Backbones Shapes Audio Language Models: A Holistic Evaluation

Large language models (LLMs) have been widely used as knowledge backbones of Large Audio Language Models (LALMs), yet how much auditory knowledge they encode through text-only pre-training and how this affects downstream performance remains unclear. We study this gap by comparing different LLMs under two text-only and one audio-grounded setting: (1) direct probing on AKB-2000, a curated benchmark testing the breadth and depth of auditory knowledge; (2) cascade evaluation, where LLMs reason over text descriptions from an audio captioner; and (3) audio-grounded evaluation, where each LLM is fine-tuned into a Large Audio Language Model (LALM) with an audio encoder. Our findings reveal that auditory knowledge varies substantially across families, and text-only results are strongly correlated with audio performance. Our work provides empirical grounding for a comprehensive understanding of LLMs in audio research.

eess.AS↗

arXiv2026

MOS-Bias: From Hidden Gender Bias to Gender-Aware Speech Quality Assessment

The Mean Opinion Score (MOS) serves as the standard metric for speech quality assessment, yet biases in human annotations remain underexplored. We conduct the first systematic analysis of gender bias in MOS, revealing that male listeners consistently assign higher scores than female listeners--a gap that is most pronounced in low-quality speech and gradually diminishes as quality improves. This quality-dependent structure proves difficult to eliminate through simple calibration. We further demonstrate that automated MOS models trained on aggregated labels exhibit predictions skewed toward male standards of perception. To address this, we propose a gender-aware model that learns gender-specific scoring patterns through abstracting binary group embeddings, thereby improving overall and gender-specific prediction accuracy. This study establishes that gender bias in MOS constitutes a systematic, learnable pattern demanding attention in equitable speech evaluation.

eess.AS↗

arXiv2026

SAKE: Towards Editing Auditory Attribute Knowledge of Large Audio-Language Models

Knowledge editing enables targeted updates without retraining, but prior work focuses on textual or visual facts, leaving abstract auditory perceptual knowledge underexplored. We introduce SAKE, the first benchmark for editing perceptual auditory attribute knowledge in large audio-language models (LALMs), which requires modifying acoustic generalization rather than isolated facts. We evaluate eight diverse editing methods on three LALMs across reliability, generality, locality, and portability, under single and sequential edits. Results show that most methods enforce edits reliably but struggle with auditory generalization, intra-attribute locality, and multimodal knowledge propagation, and often exhibit forgetting or degeneration in sequential editing. Additionally, fine-tuning the modality connector emerges as a more robust and balanced baseline compared with directly editing the LLM backbones. SAKE reveals key limitations of current methods and provides a foundation for developing auditory-specific LALM editing techniques.

cs.SD↗

arXiv2026

Nudging Hidden States: Training-Free Model Steering for Chain-of-Thought Reasoning in Large Audio-Language Models

Chain-of-thought (CoT) prompting has been extended to large audio-language models (LALMs) to elicit reasoning, yet enhancing its effectiveness without training remains challenging. We study inference-time model steering as a training-free approach to improve LALM reasoning. We introduce three strategies using diverse information sources and evaluate them across four LALMs and four benchmarks. Results show general accuracy gains up to 4.4% over CoT prompting. Notably, we identify a cross-modal transfer where steering vectors derived from few text samples effectively guide speech-based reasoning, demonstrating high data efficiency. We also examine hyperparameter sensitivity to understand the robustness of these approaches. Our findings position model steering as a practical direction for strengthening LALM reasoning.

cs.SD↗

arXiv2026

TASTE-Streaming: Towards Streamable Text-Aligned Speech Tokenization and Embedding for Spoken Language Modeling

Text-speech joint spoken language modeling (SLM) aims at natural and intelligent speech-based interactions, but developing such a system may suffer from modality mismatch: speech unit sequences are much longer than text tokens. Prior work reduces this gap with text-aligned tokenization and embedding (TASTE), producing speech tokens that align in lengths with their textual counterparts. However, the dependence on an external ASR system and the use of a non-causal decoder limits streaming use. To address this limitation, we propose TASTE-S, a streamable extension of TASTE suitable for real-time usage. TASTE-S integrates a CTC-based ASR module into the encoder for instant dual-modality encoding. We also redesign the unit decoder to enable on-the-fly decoding. With joint training, we show that TASTE-S matches TASTE's performance while significantly reducing latency. Further investigations reveal that TASTE-S remains robust to transcriptions and enables long-form encoding and decoding.

cs.CL↗

arXiv2026

Generalized Stock Price Prediction for Multiple Stocks Combined with News Fusion

Predicting stock prices presents challenges in financial forecasting. While traditional approaches such as ARIMA and RNNs are prevalent, recent developments in Large Language Models (LLMs) offer alternative methodologies. This paper introduces an approach that integrates LLMs with daily financial news for stock price prediction. To address the challenge of processing news data and identifying relevant content, we utilize stock name embeddings within attention mechanisms. Specifically, we encode news articles using a pre-trained LLM and implement three attention-based pooling techniques -- self-attentive, cross-attentive, and position-aware self-attentive pooling -- to filter news based on stock relevance. The filtered news embeddings, combined with historical stock prices, serve as inputs to the prediction model. Unlike prior studies that focus on individual stocks, our method trains a single generalized model applicable across multiple stocks. Experimental results demonstrate a 7.11% reduction in Mean Absolute Error (MAE) compared to the baseline, indicating the utility of stock name embeddings for news filtering and price forecasting within a generalized framework.

q-fin.ST↗

arXiv2026

ParaS2S: Benchmarking and Aligning Spoken Language Models for Paralinguistic-aware Speech-to-Speech Interaction

Speech-to-Speech (S2S) models have shown promising dialogue capabilities, but their ability to handle paralinguistic cues - such as emotion, tone, and speaker attributes - and to respond appropriately in both content and style remains under-explored. Progress is further hindered by the scarcity of high-quality and expressive demonstrations. To address this, we introduce a new reinforcement learning (RL) framework for paralinguistic-aware S2S, ParaS2S, which evaluates and optimizes both response content and speaking style directly at the waveform level. We first construct ParaS2SBench, a benchmark that evaluates the naturalness of input-output pairs in terms of content and speaking style using expressive and challenging queries. For the automatic judge, we propose a PolyTone training strategy and a multi-stage framework, preventing the style hallucination of end-to-end audio LLM judging. Our judge correlates well with human preferences and is scalable, enabling the model to interact and learn from unlabeled speech via RL. Experiments show that existing S2S models fail to respond appropriately to paralinguistic attributes, performing no better than pipeline-based baselines. Our RL approach (ParaS2SAlign) achieves a 10% relative improvement in the appropriateness of response content and speaking style on ParaS2SBench over supervised fine-tuning (SFT), surpassing all prior models while requiring substantially fewer paired demonstrations than pure SFT. Our findings highlight the need for a scalable and accurate automatic evaluator for speech-to-speech interaction.

eess.AS↗

arXiv2026

STITCH: Simultaneous Thinking and Talking with Chunked Reasoning for Spoken Language Models

Spoken Language Models (SLMs) are designed to take speech inputs and produce spoken responses. However, current SLMs lack the ability to perform an internal, unspoken thinking process before responding. In contrast, humans typically engage in complex mental reasoning internally, enabling them to communicate ideas clearly and concisely. Thus, integrating an unspoken thought process into SLMs is highly desirable. While naively generating a complete chain-of-thought (CoT) reasoning before starting to talk can enable thinking for SLMs, this induces additional latency for the speech response, as the CoT reasoning can be arbitrarily long. To solve this issue, we propose Stitch, a novel generation method that alternates between the generation of unspoken reasoning chunks and spoken response chunks. Since the audio duration of a chunk of spoken response is much longer than the time to generate the tokens in a chunk of spoken response, we use the remaining free time to generate the unspoken reasoning tokens. When a chunk of audio is played to the user, the model continues to generate the next unspoken reasoning chunk, achieving simultaneous thinking and talking. Remarkably, Stitch matches the latency of baselines that cannot generate unspoken CoT by design while outperforming those baselines by 15% on math reasoning datasets; Stitch also performs equally well on non-reasoning datasets as those baseline models. Some animations and demonstrations are on the project page: https://d223302.github.io/STITCH.

cs.CL↗

arXiv2026

TASTE: Text-Aligned Speech Tokenization and Embedding for Spoken Language Modeling

Recent efforts target spoken language models (SLMs) that not only listen but also speak for more natural human-LLM interaction. Joint speech-text modeling is a promising direction to achieve this. However, the effectiveness of recent speech tokens for joint modeling remains underexplored. To address this, we introduce Text-Aligned Speech Tokenization and Embedding (TASTE), a method that directly addresses the modality gap by aligning speech token with the corresponding text transcription during the tokenization stage. We propose a method that can achieve this through a attention-based aggregation mechanism and with speech reconstruction as the training objective. We conduct extensive experiments and show that TASTE can preserve essential paralinguistic information while dramatically reducing the token sequence length. With TASTE, we perform straightforward joint spoken language modeling by using Low-Rank Adaptation on the pre-trained text LLM. Experimental results show that TASTE-based SLMs perform comparable to previous work on SALMON and StoryCloze; while significantly outperform other pre-trained SLMs on speech continuation across subjective and objective evaluations. To our knowledge, TASTE is the first end-to-end approach that utilizes a reconstruction objective to automatically learn a text-aligned speech tokenization and embedding suitable for spoken language modeling. Our demo, code, and model are available at https://mtkresearch.github.io/TASTE-SpokenLM.github.io.

cs.CL↗

arXiv2026

The ICASSP 2026 HumDial Challenge: Benchmarking Human-like Spoken Dialogue Systems in the LLM Era

Driven by the rapid advancement of Large Language Models (LLMs), particularly Audio-LLMs and Omni-models, spoken dialogue systems have evolved significantly, progressively narrowing the gap between human-machine and human-human interactions. Achieving truly ``human-like'' communication necessitates a dual capability: emotional intelligence to perceive and resonate with users' emotional states, and robust interaction mechanisms to navigate the dynamic, natural flow of conversation, such as real-time turn-taking. Therefore, we launched the first Human-like Spoken Dialogue Systems Challenge (HumDial) at ICASSP 2026 to benchmark these dual capabilities. Anchored by a sizable dataset derived from authentic human conversations, this initiative establishes a fair evaluation platform across two tracks: (1) Emotional Intelligence, targeting long-term emotion understanding and empathetic generation; and (2) Full-Duplex Interaction, systematically evaluating real-time decision-making under `` listening-while-speaking'' conditions. This paper summarizes the dataset, track configurations, and the final results.

cs.SD↗

arXiv2026

AQAScore: Evaluating Semantic Alignment in Text-to-Audio Generation via Audio Question Answering

Although text-to-audio generation has made remarkable progress in realism and diversity, the development of evaluation metrics has not kept pace. Widely-adopted approaches, typically based on embedding similarity like CLAPScore, effectively measure general relevance but remain limited in fine-grained semantic alignment and compositional reasoning. To address this, we introduce AQAScore, a backbone-agnostic evaluation framework that leverages the reasoning capabilities of audio-aware large language models (ALLMs). AQAScore reformulates assessment as a probabilistic semantic verification task; rather than relying on open-ended text generation, it estimates alignment by computing the exact log-probability of a "Yes" answer to targeted semantic queries. We evaluate AQAScore across multiple benchmarks, including human-rated relevance, pairwise comparison, and compositional reasoning tasks. Experimental results show that AQAScore consistently achieves higher correlation with human judgments than similarity-based metrics and generative prompting baselines, showing its effectiveness in capturing subtle semantic inconsistencies and scaling with the capability of underlying ALLMs.

eess.AS↗