Papers

In cross-lingual zero-shot text-to-speech, the accent of the reference leaks into the target speech. We propose accent analogy guidance (AAG), a training-free sampler term that subtracts an accent direction estimated from the model's own predictions for one synthetic voice rendered in both languages, so the voice cancels and only the accent remains. By a blind LLM accent judge on real dubbing data, reweighting classifier-free guidance between reference and text, and its variants, stay near one identity-accent trade-off curve; we score a method by its speaker similarity above that curve at equal accent ($\Delta$SIM). Across four open TTS models AAG lies above the curve: on OmniVoice $\Delta$SIM is +0.11 to +0.27 on three test sets (accent 3.51 to 4.28 on a 1-5 scale at speaker similarity 0.29, where reweighting keeps 0.02); MaskGCT and CosyVoice 2 also lie above their curves, and on F5-TTS it is more native than any reweighting setting. An LLM-free language-ID measure and a twelve-listener panel agree. A premise test and the reach of a model's own curve indicate in advance whether and roughly how much AAG can gain, predicting the one model where it gains nothing (X-Voice).

The constant-Q transform (CQT) represents audio on a logarithmic frequency axis. Its nonstationary Gabor formulation is exactly invertible, but the unequal numbers of time coefficients in its bands complicate GPU computation. An exact factorisation combines spectral selection, conjugation, windowing, and reordering into a fixed map between one packed Fourier transform and the shorter band inverse transforms. The factors give waveform reconstruction, real adjoints for backpropagation, and bounds on arithmetic depth and block width; overlapping slices permit streaming with bounded memory. Tests on two GPU models show that Flash-CQT reduces analysis-synthesis round-trip time by factors of two to eight relative to a baseline computing the same CQT. The proposed implementation also uses over 30% less peak temporary workspace and reaches a negligible reconstruction error, with a signal-to-noise ratio of about 130 dB, in single-precision floating-point arithmetic. These advances make Flash-CQT a practical, computationally efficient front end for spectral analysis and modern audio machine-learning systems.

Score-informed note separation seeks to extract the performed waveform of all individual notes, often from a polyphonic recording. Existing deep learning systems generally only target instrument-level stems. We present, to our knowledge, the first deep learning approach to score-informed note separation, NoteSep. NoteSep extracts the queried notes by applying an extraction stage model, NoteGrab, once per note. Conditioned on pitch, onset, and offset, NoteGrab separates harmonic and percussive components in two U-Nets linked by bidirectional cross-attention; selective harmonic gating suppresses lower-octave interference while preserving percussive attacks. Finally, a joint separation stage applies Adaptive Set Ownership (ASO) to compare concurrent NoteGrab estimates and reallocate mixture energy. We curate SCNS-Train (25,729 mixtures and 743,920 targets) for training and SCNS-Eval (16 instruments, disjoint scores and libraries) for evaluation. On SCNS-Eval, NoteSep reaches a median SI-SDR of 7.39~dB, compared with 2.49~dB for our strongest baseline. See the demo page at https://benschou.com/notesep.

Neural audio watermarks are increasingly deployed in commercial speech generation systems to make AI-generated speech traceable, yet their robustness has been studied mainly under conventional signal distortions. Since a watermark can be regarded as imperceptible noise added to the speech signal, a natural question is whether speech enhancement (SE), as a denoising model, can remove it. In this paper, we cascade Gaussian noise with SE models as a black-box watermark removal attack, covering both discriminative and generative SE paradigms, against six neural watermarks: AudioSeal, WavMark, SilentCipher, Timbre, Perth, and AlignMark. Experimental results show that the proposed attack significantly outperforms existing neural re-synthesis methods in watermark removal. In particular, we find that generative SE, which reconstructs the harmonic regions of speech while denoising, is highly destructive to watermarks. These findings show that SE poses a serious threat to current audio watermarking methods, and we call for SE-aware robustness evaluation in watermark design.

Post-training quantization (PTQ) reduces the cost of on-device text-to-speech (TTS), but published evaluations cover one system or method. We evaluate PTQ across TTS architectures under one protocol with three core models, weight and activation ablations of eight more, and two held-out models quantized blind. Four-bit per-channel weights reduce UTMOS, a predicted mean opinion score, by 2.8 on Supertonic and 0.07 on Kokoro, and per-tensor scaling can cause severe degradation even at 8 bits. The same bit width yields different outcomes, because the sensitive component is model-specific and not reliably predicted from the model class. A staged ablation procedure identifies it, and per-layer GPTQ can restore it to within 0.1 UTMOS. Real int8 and int4 kernels reproduce the simulated ordering at hardware-dependent cost. On a Mac mini, a 4-bit weight kernel runs Supertonic at 0.60x the fp32 latency while int8 is slower, so each configuration requires validation on the target runtime.

Audio carries rich cues about human activities, and microphones are already built into most wearable devices. However, microphones also capture speech, and this privacy risk limits their use in Human Activity Recognition (HAR). We present AnomaSense, a sensor activation approach for wrist wearables that keeps the microphone off by default and turns it on for at most one second when an unsupervised anomaly detector flags an IMU segment that is likely to produce sound. The captured audio is further masked before it reaches the recognition model. We study 20 activities from 15 participants, organized into five groups in which activities share similar wrist motion but differ in the object or material involved. With IMU data alone, our recognition model reaches 78.98% accuracy in leave-one-participant-out validation. With the short, masked audio windows added, accuracy reaches 96.89% with no masking and stays above 86% when 90% of each one-second audio window is removed. On the same data, the anomaly detector triggers the microphone with 86.46% precision and 74.28% recall relative to sound events. We also report a small preliminary check of automatic speech recognition on masked speech, which shows that contiguous masking degrades recognition far more than point-wise masking at the same masking ratio. Our evaluation is a controlled, offline feasibility study. We describe the threat model, what the approach does and does not protect, and the steps needed before deployment.

In flow-matching text-to-speech (TTS), different speech realizations can induce different target velocities under the same generation conditions. A deterministic velocity field trained with squared error predicts their conditional mean, thereby marginalizing realization-dependent variation. Meanwhile, modeling such variation does not inherently provide a semantically interpretable interface for attribute manipulation. We propose ReaFlow-TTS, a realization-conditioned flow-matching framework that introduces an utterance-level stochastic realization latent and uses it to condition velocity prediction throughout the generation trajectory. We further impose valence-arousal-dominance (VAD) semantics on the realization space, enabling direct and graded attribute manipulation without target speech at inference. Experiments demonstrate improved synthesis quality over a matched full-mask baseline and reproducible latent-induced pitch, energy, and timing tendencies across initial-noise samples, providing behavioral evidence that the latent is used as a reusable realization condition. Subjective evaluation further demonstrates graded VAD manipulation across generation contexts with only modest changes in naturalness.

Speech and voice are multidimensional signals that capture both communicative intent and underlying physiological processes, providing a unique, non-invasive window into health. Analyzing these signals has the potential to yield digital biomarkers that (i) provide scalable, objective measurement tools for research and clinical care and (ii) reflect the presence or progression of diverse conditions, including neurological, psychiatric, respiratory, and cardiovascular disorders. Realizing this promise, however, requires the field to overcome pervasive reproducibility and generalizability issues due to heterogeneous data collection, processing, and analysis practices. A major source of this heterogeneity is how underlying acoustic measures themselves are defined and computed. In this paper, we outline key considerations across the speech biomarker discovery lifecycle, from data collection through machine learning modeling to clinical interpretation, needed to achieve reliable, reproducible, and clinically translatable results. Chief among these is the need for harmonization efforts to start from common, precisely specified measure definitions. As a first step, we therefore provide definitions, physiological correlates, and computational implementations for a minimal, clinically interpretable set of core speech measures spanning respiration, phonation, articulation, and fluency. We close by discussing ongoing standardization efforts and the open challenges that remain in advancing the adoption of speech- and voice-based digital biomarkers.

Audio-language models can produce confident answers unsupported by the audio, motivating uncertainty estimates that identify unreliable responses. We compare probability-based, sampling-based, self-verification, evidential, and contrastive measures across four open-weight models and five audio QA benchmarks. In multiple-choice evaluation, first-token measures are strongest overall, with top-1 probability achieving a mean AUROC of .740, compared with .708 for ten-sample discrete semantic entropy, while requiring no additional model calls. Across four benchmarks, shifting from multiple-choice to open-ended evaluation lowers mean accuracy from 57.6% to 36.6%, yet uncertainty remains predictive of errors: semantic entropy, maximum token entropy, and semantic agreement achieve mean AUROCs of .697, .694, and .693, respectively. To test whether uncertainty reflects the evidence available to answer the question, we perform input ablations that remove either the audio or the question. Across top-1 confidence, entropy, and sampling-based measures, removing audio reduces error-detection AUROC by .101 on average, compared with .010 when removing the question. Together, these results establish efficient uncertainty baselines and show that uncertainty in audio-language models depends substantially more on available audio evidence than on question text.

Domain Generalization (DG) aims to develop models that remain robust to conditions unseen during training. While DG has been systematically studied in computer vision through controlled benchmarks and diverse distribution shifts, its evaluation in spoken language recognition remains less structured. Existing speech datasets provide valuable benchmarks for robustness to real-world acoustic conditions, but are primarily designed around specific scenarios and large scale rather than as general-purpose tools for systematically constructing and evaluating domain shifts. In this work, we introduce a smallscale speech dataset and evaluation protocol for controlled studies of acoustic domain shifts. It enables the evaluation of spoken language recognition models under a variety of acousitc domain shifts. We introduce the speech modality into the DomainBed Domain Generalization framework and evaluate three domain generalization algorithms. We show that in-domain performance is not a reliable predictor of cross-domain robustness and verify that explicit domain invariant algorithms such as MMD or DANN algorithms do not outperform ERM. We further validate the generality of these findings on MMS-LID-126, a state-of-the-art spoken language identification system. We release the code.

Machine learning tools have significantly aided automatic piano music transcription; however, this domain has focused primarily on accurately predicting the pitches and timings of played notes. To produce sheet music for the piano, notes must be separated into two staves, one for each hand, and good sheet music often contains fingering annotations to guide the player when sight-reading or learning fast or complex pieces. We propose 8 statistical approaches for combined hand and fingering annotation of transcribed piano notes, including baseline hidden Markov models, rule-based methods, a synthesis of existing approaches, and N-gram language models. Furthermore, we develop a pipeline for complete transcription from piano audio to fingering-annotated sheet music. Evaluations with the PIG dataset demonstrate that our Synthesis model achieves a hand separation accuracy of 90.8% and a joint hand and finger annotation accuracy of 56.6%. These approaches serve as a new baseline for further research into this problem, while our pipeline demonstrates the feasibility of a combined system for automated note transcription, hand separation, and fingering annotation.

Audio-language models (ALMs) can exploit textual shortcuts to answer questions while overlooking acoustic evidence, weakening audio understanding. On-policy distillation (OPD) trains compact ALMs by supervising student-generated responses with teacher predictions, but does not explicitly distinguish acoustic support from linguistic predictability. We propose Reward-Tilted On-Policy Distillation (RT-OPD) to strengthen acoustic grounding. Given the same question and student-generated text, a frozen teacher predicts the next token with and without audio inputs. Their log-probability contrast defines a reward that reshapes the teacher distribution for reverse-KL distillation, emphasizing the additional evidence provided by audio. Across two compact students and three benchmarks, RT-OPD consistently outperforms Vanilla OPD. Experiments with silenced and replacement audio further suggest that RT-OPD strengthens the student's reliance on acoustic evidence. Our 3B model achieves 72.72% accuracy on MMAU, the highest among the compared 3B models and competitive with several 7B and 8B models. Code and model checkpoints are available at https://github.com/KaiyangLi1992/RT-OPD.

We describe our submission to the Unsupervised Speech in the Wild (UPS) Challenge at Interspeech 2026, a bidirectional Mamba-2 (BiMamba2) encoder trained with masked discrete-unit prediction following the HuBERT-style paradigm. The 47.88M-parameter model is trained on 250 hours of speech across 67 languages from the MLCommons Unsupervised People's Speech dataset, with no labeled data. The objective combines masked k-means pseudo-label prediction with language identification supervision and VICReg regularization. On official evaluation, the system achieves an Adjusted Rand Index of 0.735, exceeding four baselines on speaker clustering. Language identification macro-F1 (0.073) and character error rate (0.870) remain below supervised baselines. We analyze a local-official discrepancy in metric scale and checkpoint ranking, highlighting limitations of in-distribution diagnostics for predicting Dynabench probe outcomes.

Automatic speech recognition models are audited for demographic fairness at full precision, yet the models that ship to production have been quantized, pruned, and distilled. We ask whether post-training weight compression, which alters model weights rather than the audio signal or its feature representation, redistributes error burden across demographic groups. Across the Whisper family on Fair-Speech, Common Voice 25, and AfriSpeech-200, 50% Wanda pruning of Whisper-large-v3 sharply widens the Black/AA-vs-Asian temporal-taxation differential on Fair-Speech: the absolute word-error-rate gap between the worst- and best-served groups more than doubles; at an assumed cost of five seconds of correction effort per transcription error this is a rise from 30 to 64 seconds of correction time per minute of speech. This +111% relative increase is invariant to the assumed per-error cost, survives an audio-quality control, and is only partly mitigated by beam-search decoding, which still leaves an +86% increase. At edge model size, INT4 HQQ quantization compounds catastrophic transcript loops on West African accents by factors of five to seven. Distillation, by contrast, narrows demographic gaps in 21 of 27 evaluated settings (teacher-student pair, precision, and dataset), with the exceptions concentrated on a single model pair. We cast the temporal-taxation construct of Choi and Choi (2025) as a quantitative metric, and show that single-snapshot fairness audits on full-precision models do not capture the deployment-time burden that compression places on already-marginalized speakers.

Contextual biasing improves rare-word recognition in speech large language models (SpeechLLMs), but efficiently exploiting large bias lists remains challenging. We propose PTC-Bias, a two-stage framework based on phoneme-level temporal competition. At the prefill stage, PTC Retrieval performs frame-synchronous phoneme decoding and temporal competition among candidate pronunciations, producing a compact bias-word shortlist and corresponding speech intervals. After SpeechLLM decoding, PTC Correction conducts a second local competition between the retrieved candidates and mismatched transcript spans within these intervals. Selective correction reduces near-homophone and word-segmentation errors while preserving correct transcriptions. Both stages share the same phoneme posteriors and require no additional SpeechLLM forward pass. Experiments on LibriSpeech show consistent gains across two SpeechLLMs and bias lists of up to 2000 words. With Prompt-SLAM-ASR-7B and 2000 bias words, PTC-Bias reduces B-WER by 23.4%/23.9% relative to CTC-Filter on test-clean/test-other, while keeping U-WER nearly unchanged.

We present the first phoneme-level intelligibility evaluation of speech anonymizers, assessing the performance of ASR-ensemble-based metrics against measured intelligibility from a crowdsourced listening test. Our results show that simple hard-voting ASR metric reaches correlations above 0.9 with human ratings when aggregated by feature, test-type, or condition, provided that multiple ASR models are combined; evaluating stimuli with and without a carrier sentence further improves the correlation at the stimulus level. However, posterior-probability-based confidence metrics bring no gain, which can be traced back to the insufficient calibration of the state-of-the-art open ASR models that were utilized here. All data, code, and evaluation tools are released as open source.

Audio-language models (ALMs) integrate acoustic perception with the knowledge encoded in language models, enabling contextual understanding of auditory events. Making these capabilities practical on devices with limited memory and computation motivates our focus on small ALMs with fewer than 200M parameters. We introduce a recipe that brings together architecture, data, and three-stage training to build Mizar, a 159.3M-parameter ALM. Its architecture connects a compact CED-Small audio encoder to SmolLM2-135M through a frequency-merging mapper. With supervision drawn from ReasonAQA, AudioMCQ, and AVQA, the model undergoes three training stages: audio-language alignment (Stage 1), audio-dependent fine-tuning (Stage 2), and post-training (Stage 3) aimed at strengthening weak skills while retaining learned capabilities. Across five random seeds, Mizar achieves mean accuracies of 52.92% on MMAU, 42.42% on MMAR, and 36.02% on ADQA-clean, surpassing the previous best-performing ALM below 200M parameters on all three benchmarks. It also supports local inference on a single CPU: on questions from the MMAU benchmark, the mean latency from opening the audio file to generating a complete answer is 1.09 seconds. Code and checkpoints are available at https://github.com/KaiyangLi1992/Mizar_159M.

Text-to-audio (TTA) generation aims to synthesize realistic audio that faithfully reflects natural-language descriptions. Most TTA systems adopt a two-stage latent paradigm: an audio tokenizer is optimized for reconstruction and then frozen, after which a generative model is trained in the resulting latent space. However, reconstruction-oriented representations may be suboptimal for generation, motivating joint representation and generative learning. To this end, we introduce Unite-Audio, to our knowledge, is the first to jointly learn continuous audio representations and latent flow matching for TTA. By coupling reconstruction with self-supervised generative prediction, Unite-Audio allows the generative objective to directly shape the latent space rather than treating it as a fixed intermediate representation. We further employ Flow-GRPO post-training to improve text-conditioned generation. Experiments show competitive TTA performance with a compact latent flow model, while ablation studies confirm the benefit of jointly learning the audio representation and generative model. Audio samples are available at https://runwushi.github.io/Unite-Audio.

Automatic pitch correction (APC) requires distinguishing the unintended intonation errors from expressive pitch variation. Existing systems either lack explicit harmonic modeling, as vocal-only methods do, or do not directly use the note-level polyphonic context. Therefore, we propose MIDIBack, a note-level APC framework that jointly models the vocal and accompaniment events in a shared OctupleMIDI sequence. We evaluate MIDIBack under 6 note corruption regimes, including global outshift, learned note-dependent detuning, uniform perturbations, and their combinations. The resulting model achieves 78.6% overall raw pitch accuracy (RPA), and 81.5% under combined global outshift and learned detuning. Removing the accompaniment conditioning reduces RPA from 81.5% to 35.8% in outshift, showing the effectiveness of accompaniment context. Case studies on accompaniment modulation further illustrate that vocal note predictions

In multimodal emotion recognition (MER), human affective states are inferred by integrating complementary cues from multiple modalities. In audio-text MER, affective cues are often entangled with speaker style and lexical content, while cross-modal disagreement further complicates how the evidence should be integrated. Under conventional discriminative fusion, multimodal evidence is compressed into a terminal prediction, with modality-specific cues and conflict information insufficiently preserved. In large generative affective models, by contrast, affective reasoning is typically embedded in language decoding, leaving emotion evidence implicit and difficult to verify in a structured space. To address these limitations, BiCFlow-MER (Bidirectional Conditional Flow for Multimodal Emotion Recognition) is proposed as a conditional-flow framework in which audio-text MER is formulated as generative evidence transport within a structured emotion space. Within BiCFlow-MER, emotion-oriented evidence is disentangled from speaker-style and lexical-content factors to construct a conflict-aware affective condition. Guided by this condition, each utterance is transported to an explicit emotion-space endpoint through a bidirectional rectified flow. Candidate emotions are jointly verified through adaptive prototype-cloud scoring of the transported endpoint and backward class-to-condition consistency with the original multimodal condition, enabling conflict-aware recognition. BiCFlow-MER is shown to outperform all compared methods across IEMOCAP, MELD, and the zero-shot CASE benchmark. By orchestrating discriminative recognition and generative evidence modeling through conditional transport, BiCFlow-MER defines a new MER paradigm.

This paper summarizes the Interspeech2026 second Multilingual Conversational Speech Language Model (MLC-SLM) Challenge, which aims to advance the development of effective multilingual conversational speech language models. We describe the two challenge tasks: multilingual conversational speech diarization and recognition, and multilingual conversational speech understanding, together with the released real-world conversational speech dataset, evaluation protocols, and baseline systems. The challenge attracted 91 teams worldwide, with 704 valid leaderboard results and 14 technical reports across the two tasks. Based on the participating systems, we summarize representative approaches and distill practical insights into multilingual conversational speech recognition and understanding to support future research in the community.

This paper introduces Passing, an interactive audiovisual installation that generates an endless journey from a single continuous monorail-window recording by reconstructing it as a spatiotemporal volume. Rather than replaying the footage linearly, the work resamples its spatial and temporal structure along nonlinear trajectories, producing a continuously passing landscape whose depth, speed, and temporal order become unstable. A camera-based viewer-presence detection system estimates whether a viewer is present in the viewing zone and uses this presence state to influence transitions among rendered video sequences. The resulting video stream is fed into SpecMaskFoley, a real-time video-to-audio synthesis model that generates a synchronized soundscape for the reconfigured image. The model is not used to reconstruct an objectively correct soundtrack, but functions as a speculative listener, proposing a possible auditory interpretation of a world whose conventional spatial and temporal premises have been disrupted. Passing distributes creative agency across the artist, who defines the rules of spacetime reconstruction; the AI model, which interprets the emergent visual flow as sound; and the audience, whose embodied presence influences the audiovisual trajectory. Through this structure, the work investigates how authorship and listening may be negotiated among human intention, machine inference, and audience interpretation. Artwork page: https://ryufurusawa.com/passing

Recent zero-shot text-to-speech (ZS-TTS) systems can reproduce a speaker's voice with high fidelity from only a few seconds of reference speech, raising concerns over unauthorized voice cloning and impersonation. Speaker identity unlearning has recently emerged as an approach to selectively suppress this capability for speakers who opt out while preserving synthesis capability for other speakers. Although existing approaches reduce speaker similarity, preventing re-identification often faces severe degradation of speech quality. Motivated by this observation, we propose GUARD, a lightweight speaker identity unlearning framework that combines a learned speaker gate with speaker-agnostic activation steering on a frozen TTS backbone. The steering vectors are optimized using group-relative reward optimization to shift outputs from forget speakers toward population-level impostor similarity while preserving intelligibility and speech naturalness. On CosyVoice2, GUARD reduces forget-speaker similarity from 0.541 to 0.103 and re-identification accuracy in a 150-speaker gallery from 73.5% to 0.5%, while preserving retain-speaker reproduction. The results demonstrate that similarity reduction alone may not fully characterize successful speaker identity unlearning and highlight re-identification as a complementary criterion for its evaluation.

Audio language models understand what is said far better than how it sounds. Closing this gap takes more than data. Detailed acoustic annotation is costly, labels from stronger models inherit their errors and limits, and fixed data cannot adapt as the learner improves. We therefore propose EvoAudio, a recursive self-improvement system for audio understanding. To our knowledge, it is the first to evolve the model, waveforms, questions, and difficulty in one closed loop. EvoAudio uses the current model's performance to set the focus and difficulty of the next training data. A library of audio tools then constructs questions whose answers follow from how the audio was made, providing verifiable supervision without new human annotation. Reinforcement learning updates the model, and validation decides whether it enters the next evolution round. Across 13 rounds, EvoAudio improves five models with different audio encoders and language backbones on MMSU, MMAU-Pro, and MMAR. It achieves the highest average for every backbone, raising overall performance by up to 6.3 points. The improvement unfolds over successive rounds, with each stronger model starting the next round.

Speech technology penalizes some voices: recognition errs nearly twice as often for Black speakers, and accuracy declines for second-language accents and older speakers. We introduce TRIAD, an audit grid crossing 120 texts, 24 rendered demographic voice profiles (gender, age band, accent), and ten expressive styles via controllable text-to-speech, isolating perceived demographic attributes from content and affect. For ten open-weights encoders we define axis-fidelity functionals, principal-angle leakage between axis subspaces, and group-conditional gaps; a proposition proves that average probe disparity grows with the same aggregate voice-semantic leakage $\Lambda$ we measure, and a corollary shows that peak leakage forces worst-case disparity inside an active region. The measured mean-square probe disparity tracks $\Lambda$ (Pearson r = 0.93), and a black-box protocol exposes the same signature in two closed-source models. ORCA, an adapter combining axis-specific contrastive heads, an orthogonality penalty, and group-balanced sampling, cuts leakage 72% and roughly halves the gaps.

End-to-end speech-to-speech dialogue models listen and speak simultaneously, so a continuously open acoustic channel is exposed to adversarial manipulation. We formalize imperceptible attacks on full-duplex agents as optimization over additive perturbations confined beneath the psychoacoustic masking threshold of the carrier speech, under three goals: targeted semantic hijacking, response suppression, and policy jailbreaking. Against an undefended Moshi-style agent, white-box attacks succeed in up to 91.7% of trials. We then introduce psychoacoustically aligned latent smoothing (PALS), which injects anisotropic Gaussian noise shaped by local codebook covariance at the residual-vector-quantized latent interface, with input noise shaped by the masking threshold constraining the attacker and trained by a Kullback–Leibler consistency objective. Deployed with no inference-time cost, PALS reduces hijack to 8.3%, mute to 11.2%, and jailbreak to 9.1% at clean quality within 2.3%. A Monte Carlo-smoothed variant certifies an ellipsoidal latent radius up to 0.616, a guaranteed floor that the empirical robustness far exceeds.

Benchmarks for full-duplex spoken dialogue models score turn-taking with binary fixed-window rules that reward immediate response or silence by completeness of the prior turn. We argue that the appropriateness of a response offset, whether delayed silence or anticipatory overlap, is conditional on the speaker's latent intent, identifiable only from that speaker's behavior. We introduce TACT, a benchmark of 9,728 episodes and 73.2 hours from five dyadic corpora; each episode carries dialogue history, a per-speaker memory profile, and an annotator-derived posterior over six intent classes. Scoring replaces binary windows with a strictly proper threshold-weighted continuous ranked probability score whose weights are intent-conditioned timing kernels fitted to human floor-transfer-offset distributions, proving boundedness, consistency, and binary reduction. Across eleven systems the best model reaches 0.47 against a human topline of 0.86, is nearly invariant to speaker profiles, and TACT agrees with human judgments at Spearman 0.81 versus 0.46 for binary metrics.

Pneumonia is difficult to diagnose in older long-term care residents; multimorbidity and atypical presentations obscure signs, motivating operationally efficient objective testing. We analyzed multi-channel digital stethoscope recordings from 185 Japanese residents (73 pneumonia, 112 symptomatic without), using radiologist-confirmed chest X-rays and clinician diagnoses as supervisory signals that train convolutional neural networks, multimodal fusion, and channel-based variants with time-domain Grad-CAM interpretability. Models were evaluated with repeated patient-level cross-validation showing models with X-ray supervision outperformed clinician supervision (F1 0.729, accuracy 0.783 vs. F1 0.637, accuracy 0.711). Additionally, a three-channel selection protocol maintained performance (F1 0.736; accuracy 0.803), with two mid-thoracic sites ranking highest and Grad-CAM attention overlapping adventitious sounds. These findings indicate automated multi-channel lung-sound analysis can aid long-term care pneumonia diagnosis, with X-ray supervision being more reliable than clinical, and fewer channels preserving performance while lowering acquisition times.

Non-speech interference can change a speech representation without causing comparable task loss. We test eight frozen encoders on four tasks, adding non-speech sounds throughout recordings, during speech, or in pauses. Under whole-recording interference, embedding drift tracks task loss across seven sounds, with mean Spearman correlations of 0.81-0.88. Moving the same sound between speech and pauses changes this pattern. At quiet to moderate levels, pause interference produces larger drift, while speech interference usually causes greater loss on intent recognition, speaker verification and speech recognition. Emotion recognition shows a weaker placement effect. Pause interference also changes speech-frame representations beyond the injected region. Even below the estimated recording background, interference can change embeddings as much as repeated speech takes do. Drift helps rank the effects of different sounds, but larger drift does not consistently indicate greater task loss.

Compositionality, the ability to represent complex acoustic scenes as combinations of simpler sound sources, is central to auditory perception and classical additive signal models. Still, it remains unclear whether modern pre-trained audio representations internalize additive structure without compositional supervision. Existing evaluation frameworks of audio compositional reasoning largely focus on cross-modal audio-text alignment, leaving open whether audio representations themselves exhibit additive compositional structure independent of text grounding, analogous to vector arithmetic in word representations. To investigate this, we adopt a two-step diagnostic for frozen audio representations. First, we quantify linear alignment between representations and sound source labels using canonical correlation analysis. Second, we test additive compositional generalization via leave-one-combination-out reconstruction, grouping clips by exact source-label set, averaging their representations, and predicting held-out means from per-source contributions fitted only on training combinations. With larger combination holdouts, CLAP outperforms the permuted and label-overlap baselines on FSD50K, while the speech models do not outperform the label-overlap baseline. We examine representations generated by Wav2Vec2, HuBERT, and CLAP on FSD50K and CHiME-Home datasets. All three models show consistently higher linear correlation and more accurate leave-one-combination-out reconstructions than the permuted baselines. However, only CLAP shows large cosine similarity gains, which could be associated with its training on many kinds of audio and text. Finally, we note that all three models exhibit reconstruction residuals, revealing limits of additive compositionality such as nonlinear or non-compositional audio structure.