Speech Evaluation

The development of automated speaking assessment (ASA) is limited by the scarcity of public datasets, with most existing work relying on read-aloud speech, which limits applicability to real-world communication scenarios. In this work, we introduce OpenEnded, a corpus of English practice speech from Mandarin speakers in open-response tasks. Unlike prior open-response datasets that provide only holistic proficiency scores, OpenEnded offers utterance-level assessments of accuracy, fluency, and prosody. Approximately 10,000 utterances are collected and annotated using a hybrid framework: 1,000 are manually labeled via multi-rater scoring with discrepancy resolution to form a high-quality test set, while the remaining are pseudo-labeled by an audio language model (ALM) for training and development sets. We evaluate ALMs and existing ASA models on the OpenEnded test set and introduce VoxPA as an additional baseline. Results show that ALM-generated pseudo-labels improve training over original ALM scoring, while VoxPA achieves the best performance among all baselines.

Speech-to-speech LLMs like Moshi, and its derivative PersonaPlex, can listen and speak concurrently through full-duplex generation. However, they can begin speaking inappropriately during prolonged user silence: under digital-zero input, Moshi and PersonaPlex initiate speech in 30% and 27.5% of five-minute continuations, respectively. What causes this spurious speech? We investigate two hypotheses: either repeated sampling selects speech despite persistently low onset probabilities, or self-conditioning on nonspeech outputs causes an abrupt spike in onset probability. We find that, at every observed onset, speech probability spikes by over nine orders of magnitude in one 80-ms frame, supporting the latter hypothesis. Then, to suppress these onsets without blocking genuine responses, we ask a causal counterfactual question: is the model responding to user speech, or would its next-token distribution remain similar if the preceding user input were muted? Accordingly, we suppress onsets whose distributions change little under this intervention. Under realistic microphone noise, our method suppresses spurious onsets, while preserving genuine responses: one-sided 95% lower confidence bounds are 98.68% and 98.82% for Moshi, and 96.90% and 99.25% for PersonaPlex. Our inference-time method runs in real-time without retraining, with 95th-percentile decision time below 61 ms, within the 80-ms frame budget. Our code is available at https://github.com/KentoNishi/icassp27-spurious-onsets.

Natural-language instructions enable flexible control of synthesized speech, yet emotional TTS systems primarily model a single utterance-level affect, leaving multi-emotion control underexplored. We study two complementary multi-emotion TTS tasks: emotion trajectory, which spans several ordered affective stages, and emotion blending, in which multiple emotions coexist throughout an utterance. These tasks expose a supervision mismatch: supervised fine-tuning (SFT) does not explicitly evaluate emotion features, while single-emotion rewards provide neither structure-aware feedback for trajectory completion nor pair-aware feedback for blending. We introduce HybridEmo, a post-training framework that initializes both tasks with SFT and then aligns the speech-token policy through Group Relative Policy Optimization using a sample-aware hybrid reward. For trajectory samples, segment-aligned consistency combines average and weakest-stage evidence to preserve the correctness and completeness of prescribed stages. For blending samples, a GMM-based reward combines frame-level support from the union of target-emotion anchors in an offline emotion space with an utterance-level weaker-target margin. Both branches share an ASR reward and are routed within a unified policy. On MultiEmo-Test, HybridEmo significantly improves trajectory correctness and blending intensity, without a noticeable degradation in speaker similarity. Human evaluation prefers HybridEmo to CosyVoice 3 and EmoVoice-0.5B, with nearly balanced preferences against Qwen3-TTS.

This paper presents ECHO (Enhanced Care & Health Observer), a locally-deployable conversational health assistant for long-term chronic care management. ECHO integrates three complementary software modules developed under shared supervision as a unified system. The core module is an agentic chatbot built on a ReAct loop orchestrated via LangGraph, equipped with 17 clinical tools and a temporal knowledge graph for persistent cross-session memory; it achieves a 94.9% tool-execution pass rate across a 59-scenario benchmark with GPT-5 Mini. A two-stage hybrid safety layer intercepts all incoming queries: a rule-based layer handles explicit crisis signals and jailbreak attempts in under 1ms, while a signed graph neural network (GNN) with APPNP-style propagation classifies boundary cases by clinical intent, achieving 88.8% accuracy and 90.6% unsafe recall on a 2,537-query annotated Turkish health dataset while outperforming zero-shot LLM baselines including Llama 3.3 70B. A multimodal speech assessment module combining Whisper acoustic encoding and BERT text encoding with cross-attention fusion estimates emotion, depression, and pain, reaching a mean macro F1 of 0.652. The full system is implemented as a web application that can run entirely on consumer hardware, with no patient data transmitted to external services, supporting compliance with GDPR and KVKK.

Spoken emotional dialogue requires a model to understand a user's spoken input and generate a response that is both semantically appropriate and emotionally expressive. This is challenging because communicative intent may be stated explicitly in lexical content or conveyed more implicitly through paralinguistic cues, which can complement or diverge from the words themselves. However, two limitations constrain progress in this area: the scarcity of benchmarks that distinguish these intent expressions, and the lack of reinforcement learning objectives that jointly account for response quality and emotional expression. To address the lack of suitable benchmarks, we introduce ParaIntent, a Chinese benchmark comprising 14 intent categories with balanced explicit and implicit samples, together with a multidimensional evaluation protocol covering intent fulfillment, response quality, and emotional expression. For policy optimization, existing approaches either use a shared objective for text and speech or apply reinforcement learning to only one modality, leaving modality-specific learning signals entangled within policy optimization. Motivated by this, we propose Acoustic-Lexical Decoupled Policy Optimization (ALPO), which computes independent textual and acoustic advantages and routes them to the corresponding text and speech tokens within a unified rollout. Under identical reward functions and training budgets, ALPO improves over standard GRPO on most automatic metrics and achieves the best subjective results among the fine-tuned variants, with particularly clear gains in emotional expressiveness on both the synthetic and human-recorded test sets.

Large audio-language models (LALMs) are increasingly used as automatic judges for speech evaluation. However, high agreement with human ratings does not guarantee that their verdicts are grounded in the audio. A judge may instead rely on specialist labels or reference data supplied by the evaluation protocol itself, taking a shortcut in place of listening to the audio. In this paper, we audit such protocol-level ``shortcuts'' in LALM judges across three common deployment protocols: feature-blueprint judging, where the audio is replaced by a structured text description of acoustic features, reference-conditioned judging, and pairwise A/B comparison. Across six judges and four attributes, we find that several LALMs rely on protocol-level shortcuts. For example, in feature-blueprint judging, incorrect specialist labels reduce five judges' emotion accuracy to 0.10 or below, and in concatenated A/B comparisons, Qwen3-Omni-Thinking often picks the same slot regardless of order swaps. These results indicate that aggregate agreement can overstate the validity of LALM judges unless the model and the evaluation protocol are assessed jointly, and that each model-protocol pair should be evaluated with a matched shortcut probe.

Multimodal sentiment analysis relies on language, visual, and acoustic cues, but utterance-level modality quality may vary due to occlusion, background noise, motion blur, or imperfect transcripts, causing conventional fusion to over-trust unreliable modalities. We propose MRUF, a reliability-aware fusion method that combines multi-granularity routing with uncertainty-aware calibration. MRUF summarizes sentiment-relevant representations, performs subspace- and modality-level routing, and supervises modality routing with leave-one-out error increases to estimate utterance-level modality importance. It further predicts modality-wise uncertainty and refines modality gates through inverse-variance reweighting, while modality-invariant contrastive alignment stabilizes the shared representation space. Experiments on CMU-MOSI and CMU-MOSEI under aligned and unaligned settings show consistent improvements over strong baselines, and mechanism analysis verifies that modalities with higher predicted uncertainty receive lower fusion weights.

We report the empirical reliability of Gemini models as audio judges that score full-duplex agent conversations directly from the raw stereo waveform, tested across three models in the Gemini family: 2.5 Flash, 3.5 Flash, and 3.1 Pro. Our primary evidence base uses Gemini 2.5 Flash as the ground-truth model, validated against three calibrated human raters on 209 stereo sessions, scored on 8 production dimensions: 152 full-duplex conversations across 13 accent-and-condition strata, together with 57 adversarial defect-injected clips. The evidence for Gemini 2.5 Flash is consistent across three tests. (i) On 5 of 8 dimensions the LALM-human Spearman rho departs from the pairwise human-human rho by at most 0.07, and on 7 of 8 dimensions the two quantities 95 percent bootstrap confidence intervals overlap. (ii) The LALM agrees with the three-rater human mean within 1 point on 60 to 92 percent of sessions on 6 of 8 dimensions. (iii) On 45 of 48 (defect, dimension) cells the LALM is as sensitive as humans or better under Newcombe-Wilson 95 percent confidence intervals, though most of these are underpowered nulls rather than demonstrated parity. Rank-ordering ability transfers across the Gemini family: 3.5 Flash improves simple agreement to 8 of 8 dimensions, while 3.1 Pro rates several dimensions markedly lower than humans despite comparable rank correlation. A model swap should be re-validated on calibration specifically, not assumed from rank-correlation alone. We identify four areas where deployment requires care, and we estimate that human rating alone for our current evaluation cadence costs roughly two orders of magnitude more than the equivalent LALM workload. The data presented here provides a defensible empirical basis for deploying the LALM as a substitute or fourth rater on the dimensions where the evidence supports it.

Turn-taking naturalness is central to full-duplex spoken dialogue systems, yet its automatic evaluation remains limited. Existing evaluations often rely on human judgments or behavior-specific timing metrics, making it difficult to compare heterogeneous timing failures within a unified framework. We propose TurnNat, a likelihood-based framework for automatic turn-taking naturalness evaluation in two-channel spoken dialogue. A causal turn-taking prediction model trained on natural conversations estimates future two-speaker voice-activity states, and the negative log-likelihood (NLL) of the observed future activity measures timing atypicality. TurnNat pools frame-level NLLs over turn-taking boundary units (TBUs) extracted from utterance onsets and offsets, and aggregates mean and tail TBU scores into a dialogue-level naturalness score. We further construct a controlled perturbation benchmark of paired natural and perturbed dialogue clips, validated by human naturalness judgments. Experiments on this benchmark show that TurnNat successfully identifies unnatural turn-taking perturbations across heterogeneous timing failures.

Text-to-speech (TTS) evaluation is an open challenge. While the primary target was "naturalness," recent fidelity gains shifted focus toward "appropriateness" and whether speech is correct for its context. In this work, we examine how perception changes when the expected downstream use varies. We measure the appropriateness and human-likeness of five SOTA TTS systems across five domains: AI assistant, reader, actor, animated character, and spontaneous speaker. Results show appropriateness varies across domains independently of naturalness. While systems shine at reading, expressive domains remain challenging, and optimizing for one can degrade others. Furthermore, naturalness scores tend to penalize stylized speech while rewarding spontaneity. Finally, our study also highlights blind spots in one-size-fits-all evaluation metrics across more expressive domains. We demonstrate that TTS performance is not "solved" but depends on the target domain, requiring context-aware evaluation.

As multimodal conversational systems increasingly engage in spoken interaction, their ability to navigate paralinguistic social cues has become a critical bottleneck for natural human-AI communication. However, existing evaluations of machine emotional intelligence assess reasoning exclusively through isolated text or passive acoustic perception, overlooking the complex cross-modal reasoning required for active, multi-turn dialogue. We introduce \textsc{SpeechEQ}, a comprehensive framework designed to evaluate the sociolinguistic reasoning of Speech-Language Models (SLMs). The framework includes a validated dataset of 2,265 dialogues across 15 Emotional Quotient (EQ) subscales grounded in EQ-i 2.0 theory, along with a multi-turn evaluation protocol measured by our proposed Spoken EQ (SEQ) score inspired by human EQ assessments. Experiments show limitations in how both existing Speech Emotion Recognition and end-to-end Speech-Language Models understand and apply paralinguistic cues through speech. While end-to-end architectures outperform cascaded systems, \textsc{SpeechEQ} reveals that current multimodal models remain bottlenecked by a text-reliant modality shortcut,'' an alignment-induced safety trap,'' and ``contextual amnesia,'' highlighting the barriers to truly emotionally aware AI. Our benchmark can be accessed at https://huggingface.co/datasets/SpeechEQ/SpeechEQ and demo page at https://binomial14.github.io/speecheq-demo/

Reviewing recorded interviews for affective cues such as composure, hesitation and agitation is slow and subjective, and cloud services that could automate it require sensitive audio to leave the device. EmotionAI is a fully local Computational Intelligence (CI) pipeline that couples Speech Emotion Recognition (SER) with generative reasoning. Speaker diarisation, Whisper Automatic Speech Recognition (ASR) and a wav2vec2 emotion classifier produce per-segment affective evidence, which is then passed to an adversarial three-model local Large Language Model (LLM) panel for timestamp-grounded and citation-constrained question answering. Zero-shot evaluation on the RAVDESS four-class English subset (n = 672) exposes cross-corpus fragility rather than classifier superiority: the deployed classifier scores 48.8% accuracy, above random (24.9%) and majority (28.6%) baselines but below an in-domain MFCC + logistic-regression comparator (71.0%). The complete pipeline runs in a mean 157 s on CPU (real-time factor approximately 1.33) with zero external calls. The contribution is not state-of-the-art SER but an auditable, privacy-preserving integration of imperfect affective evidence into grounded conversational analysis, together with an honest empirical account of where cross-corpus transfer and human-centred validation still fall short.

Humans tend to speak louder and clearer in challenging environments, such as noisy conditions or when addressing hearingimpaired listeners, which is called Lombard effect. To simulate this behavior in speech synthesis systems, we introduce a flow-matching based text-to-speech (TTS) model trained with vocal effort and articulation pseudo-labels. The proposed model achieves continuous and disentangled control of vocal effort and articulation, while also enabling word-level emphasis for clarifying specific segments of an utterance. Experimental results show that these control mechanisms effectively improve clarityrelated acoustic features. Furthermore, speech-in-noise experiments demonstrate that our model successfully simulates the intelligibility gains of human clear speech in noisy conditions.

We present ProSarc, an audio-only framework that detects sarcasm by modelling temporal prosodic incongruity, that is, the mismatch between local prosodic dynamics and the utterance-level emotional baseline. Dual encoding paths, a Global Emotion Encoder and a Temporal Prosody Encoder (BiLSTM + multi-head attention), feed a Prosodic Incongruity Analyzer that produces a scalar incongruity score for classification. Monte Carlo dropout provides uncertainty estimates, and an attention-based mechanism localises sarcastic onset without frame-level labels. ProSarc outperforms prior audio-only methods on MUStARD++ (F1=75.3) and generalises to spontaneous (PodSarc, F1=62.9) and cross-lingual speech (MuSaG, F1=65.6). Ten-run validation confirms the contribution of incongruity modelling (Wilcoxon p=0.002, Cohen's d=1.51). Human evaluation shows that model uncertainty tracks perceptual ambiguity and predicted onsets align with human-annotated temporal windows.

Detecting psychological defense mechanisms in conversational text remains a challenging clinical NLP problem. For the PsyDefDetect 2026 shared task (nine-class utterance classification evaluated via macro F1), our team LinguIUTics achieves a macro F1-score of 0.3917 on the official positive-class leaderboard, ranking 4th out of 21 registered teams and improving over the Ministral-8B task baseline (31.48 macro F1) by 7.7 absolute points (24.4 percent relative). BERT-family encoders and zero-shot LLMs proved ineffective on rare classes due to severe class imbalance, leading us to QLoRA fine-tuning of Qwen3-8B. We leverage three key strategies: grouped stratified cross-validation (preventing leakage), minority-class round-robin lexical augmentation, and a post-processing pipeline with logit bias tuning and ensemble blending. Together, these components close much of the validation-to-leaderboard gap and substantially improve minority-class recall, driving the critical "Unclear" class (Level 8) from near-zero performance to an F1 score of 0.797.

Evaluating speech generation still relies heavily on human judgments, such as Mean Opinion Score (MOS), which are expensive, subjective, and difficult to reproduce at scale. While a few recent studies have begun to explore AudioLLM-based judge models, existing efforts typically target only a narrow set of scenarios (e.g., utterance-level quality or single-turn dialogue) and provide limited coverage of diverse speech generation tasks and evaluation dimensions. In this work, we propose UniSRM, a unified speech reward model that can support multi-dimensional, interpretable reward signals with reliable reasoning. To support training and evaluation, we introduce UniSRM-Data and UniSRM-Bench, covering speech evaluation tasks from utterance-level quality to context-level coherence. Based on this dataset, we present the unified speech reward model, UniSRM, with a two-stage pipeline that enables reasoning-based fine-grained assessment. Furthermore, we introduce Reasoning-Consistent Rewards to improve the reliability of the reasoning process. Experiments show that UniSRM delivers more reliable and human-aligned judgments across a broad range of speech evaluation tasks, offering a practical foundation for scalable and unified evaluation of speech quality.

We investigate whether acoustic emotion recognition models can serve as proxies for the Pathos dimension in political speech analysis, as operationalised by the TRUST multi-agent large language model (LLM) pipeline. Using a Bundestag plenary speech by Felix Banaszak (51 segments, 245 s) as a case study, we compare three analysis modalities: (1) emotion2vec_plus_large, an acoustic speech emotion recognition (SER) model whose continuous Arousal and Valence values are derived via post-hoc Russell Circumplex projection; (2) Gemini 2.5 Flash, an LLM analysing the full speech audio together with its transcript in an open-ended, context-aware fashion; and (3) TRUST-Pathos scores from a three-advocate LLM supervisor ensemble. Spearman rank correlations reveal that Gemini Valence correlates strongly with TRUST-Pathos (rho = +0.664, p < 0.001), whereas emotion2vec Valence does not (rho = +0.097, p = 0.499). We further demonstrate, via a systematic quality evaluation of the Berlin Database of Emotional Speech (EMO-DB) using Gemini in an open-ended annotation paradigm, that standard SER benchmark corpora suffer from acted speech, cultural bias, and category incompatibility. Our results suggest that LLM-based multimodal analysis captures semantically defined political emotion substantially better than acoustic models alone, while acoustic features remain informative for low-level Arousal estimation. Future work will extend this approach to video-based analysis incorporating facial expression and gaze.

CHILDES is a paramount resource for language acquisition studies -- yet computational tools for analyzing its syntactic structure remain limited. Leveraging the recent release of the UD-English-CHILDES treebank with gold-standard Universal Dependencies (UD) annotations, we train a state-of-the-art dependency parser specifically tailored to CHILDES. The parser more accurately captures syntactic patterns in child-adult interactions, outperforming widely used off-the-shelf English parsers, including SpaCy and Stanza. Alongside the parser, we also release a Part-of-Speech tagger and an utterance-level construction tagger, which together form the open-source Syntactic Parsing Toolkit for Child-Adult InTeractions (CAIT). Through a detailed error analysis and a case study tracking the distribution of syntactic constructions across developmental time in CHILDES, we demonstrate the practical utility of the toolkit for large-scale, reproducible research on language acquisition.

As audio-first agents become increasingly common in physical AI, conversational robots, and screenless wearables, audio large language models (audio-LLMs) must integrate speaker-specific understanding to support user authorization, personalization, and context-aware interaction. This requires modeling who is speaking, how the voice sounds, and how recording conditions affect speaker cues. Conventional speaker verification systems provide strong scalar scores but little linguistic evidence, while current audio-LLMs and speaker-aware language models have limited ability to organize speaker information beyond binary labels or descriptive profiles. We present SpeakerLLM, a speaker-specialized audio-LLM framework that unifies single-utterance speaker profiling, recording-condition understanding, utterance-pair speaker comparison, and evidence-organized verification reasoning within a natural-language interface. We construct verification-reasoning targets and a decision-composition policy that separate profile-level evidence from the final same-or-different decision and organize recording condition, profile evidence, and the decision into a structured trace. At its core, SpeakerLLM uses a hierarchical speaker tokenizer designed to capture multiple granularities of speaker evidence. Utterance-level speaker embeddings summarize identity and profile-level cues, whereas frame-level speaker features preserve fine-grained acoustic descriptors. Experiments show that SpeakerLLM-Base improves speaker-profile and recording-condition understanding over general audio-LLMs, while SpeakerLLM-VR preserves strong generated-verdict accuracy and produces decision traces grounded in the supervised verification reasoning schema. We will release the metadata-enriched supervision dataset and target-construction code for reproducibility.

Psychological support hotlines provide critical support for individuals experiencing mental health emergencies, yet current assessments largely rely on human operators whose judgments may vary with professional experience and are constrained by limited staffing resources. This paper proposes a large language model (LLM)-based framework for automated crisis level classification, a key indicator that supports many downstream tasks and improves the overall quality of hotline services. To better capture emotional signals in spoken conversations, we introduce a paralinguistic injection method that inserts identified non-verbal emotional cues into speech transcripts, enabling LLM-based reasoning to incorporate critical acoustic nuances. In addition, we propose a reasoning-enhanced training strategy that trains the model to generate diagnostic reasoning chains as an auxiliary task, which serves as a regulariser to improve classification performance. Combined with data augmentation, our final system achieves a macro F1-score of 0.802 and an accuracy of 0.805 on the three-class classification task under 5-fold cross-validation.

Closed-Set speaker identification aims to assign a speech utterance to one of a predefined set of enrolled speakers and requires robust modeling of speaker-specific characteristics across multiple temporal scales. While recent deep learning approaches have achieved strong performance, many existing architectures provide limited mechanisms for modeling temporal dependencies across different time scales, which can restrict the effective use of complementary short-, mid-, and long-term speaker characteristics. In this paper, we propose TARNet, a lightweight Temporal-Aware Representation Network for closed-set speaker identification. TARNet explicitly models temporal information at multiple time scales using a multi-stage temporal encoder with stage-specific dilation configurations. The resulting multi-scale representations are fused and aggregated via an Attentive Statistics Pooling (ASP) module to produce a discriminative utterance-level speaker embedding. Experiments on the VoxCeleb1 and LibriSpeech datasets show that TARNet outperforms state-of-the-art methods while maintaining competitive computational complexity, making it suitable for practical speaker identification systems. The code is publicly available at https://github.com/YassinTERRAF/TARNet.

The rapid advancement of generative audio models has outpaced the development of robust evaluation methodologies. Existing objective metrics and general multimodal large language models (MLLMs) often struggle with domain generalization, zero-shot capabilities, and instructional flexibility. To address these bottlenecks, we propose JASTIN, a generalizable, instruction-driven audio evaluation framework that formulates audio assessment as a self-instructed reasoning task. JASTIN bridges a frozen high-performance audio encoder with a fine-tuned LLM backbone via a trainable audio adapter. To ensure robust zero-shot generalization, we introduce a comprehensive instruction following data preparation pipeline, incorporating Multi-Source, Multi-Task, Multi-Calibration, and Multi-Description data. Experimental results demonstrate that JASTIN achieves state-of-the-art Pearson and Spearman correlations with human subjective ratings. It consistently outperforms general MLLMs across speech, sound, music, and out-of-domain evaluation tasks without the need for task-specific retraining.

Human languages expand vocabularies by combining existing morphemes rather than inventing arbitrary forms. Communicative efficiency shapes lexical systems at multiple levels (Gibson et al., 2019), yet morphological composition -- combining morphemes through compounding or affixation -- has rarely been modeled as a historically situated speaker choice among competing morpheme sequences, leaving unanswered why a language settles on one morpheme combination over other plausible alternatives. We ask whether a trade-off between listener recoverability and speaker production cost can predict attested compositions over contemporaneously available alternatives. Here we show, within the Rational Speech Act (RSA) framework (Frank & Goodman, 2012; Goodman & Frank, 2016) using a time-indexed lexicon constructed from Corpus of Historical American English (COHA) and Corpus of Contemporary American English (COCA), that across 4323 naturally occurring English compounds and derivations spanning 1820--2019, attested compositions are systematically ranked above unattested alternatives generated from contemporaneously available morphemes. Models integrating semantic informativeness with production cost outperform semantic-only and cost-only baselines on Mean Reciprocal Rank (MRR) and top-k accuracy (Acc@k), with the advantage of the Pragmatic Speaker model ($S_1$) over the semantic-only baseline growing as the candidate set expands, where meaning alone leaves morphological choice underdetermined. These findings suggest that lexicalization reflects a communicative trade-off between expressiveness and efficiency, extending rational accounts of communication from utterance-level choice to the internal structure of words.

Recent advances in voice cloning and text-to-speech synthesis have made partial speech manipulation - where an adversary replaces a few words within an utterance to alter its meaning while preserving the speaker's identity - an increasingly realistic threat. Existing audio deepfake detection benchmarks focus on utterance-level binary classification or single-region tampering, leaving a critical gap in detecting and localizing multiple inpainted segments whose count is unknown a priori. We address this gap with three contributions. First, we introduce MIST (Multiregion Inpainting Speech Tampering), a large-scale multilingual dataset spanning 6 languages with 1-3 independently inpainted word-level segments per utterance, generated via LLM-guided semantic replacement and neural voice cloning, with fake content constituting only 2-7% of each utterance. Second, we propose ISA (Iterative Segment Analysis), a backbone-agnostic framework that performs coarse-to-fine sliding-window classification with gap-tolerant region proposal and boundary refinement to recover all tampered regions without prior knowledge of their count. Third, we define SF1@tau, a segment-level F1 metric based on temporal IoU matching that jointly evaluates region count accuracy and localization precision. Zero-shot evaluation reveals that partial inpainting at word granularity remains unsolved by existing deepfake detectors: utterance-level classifiers trained on fully synthesized speech assign near zero fake probability to MIST utterances where only 2-7% of content is manipulated. ISA consistently outperforms non-iterative baselines in this challenging setting, and the dataset, code, and evaluation toolkit are publicly released.

There are growing concerns about the risks posed by AI companion applications designed for emotional engagement. Existing safety evaluations often rely on self-reported user data or interviews, offering limited insights into real-time dynamics. We present the first end-to-end scalable framework for controlled simulation and safety evaluation of multi-turn interactions with AI companion applications. Our framework integrates four key components: persona construction with clinical and psychometric validation, persona-specific scenario generation, scenario-driven multi-turn simulation with a dialogue refinement module that preserves persona fidelity, and harm evaluation. We apply this framework to evaluate how Replika, a widely used AI companion app, responds to high-risk user groups. We construct 9 personas representing individuals with depression, anxiety, PTSD, eating disorders, and incel identity, and collect 1,674 dialogue pairs across 25 high-risk scenarios. We combine emotion modeling and LLM-assisted utterance-and harm-level classification to analyze these exchanges. Results show that Replika exhibits a narrow emotional range dominated by curiosity and care, while frequently mirroring or normalizing unsafe content such as self-harm, disordered eating, and violent-fantasy narratives. These findings highlight how controlled persona simulations can serve as a scalable testbed for evaluating safety risks in AI companions.

Sarcasm is a form of humor where expressions convey meanings opposite to their literal interpretations. Classifying and generating sarcasm using large language models is vital for interpreting human communication. Sarcasm poses challenges for computational models, due to its nuanced nature. We introduce Sarc7, a benchmark that classifies 7 types of sarcasm: self-deprecating, brooding, deadpan, polite, obnoxious, raging, and manic by annotating entries of the MUStARD dataset. Classification was evaluated using zero-shot, few-shot, chain-of-thought (CoT), and a novel emotion-based prompting technique. We propose an emotion-based generation method developed by identifying key components of sarcasm-incongruity, shock value, and context dependency. Our classification experiments show that Gemini 2.5, using emotion-based prompting, outperforms other setups with an F1 score of 0.3664. Human evaluators preferred our emotion-based prompting, with 38.46% more successful generations than zero-shot prompting.