Audio Language Models

Siri Expressive Voices synthesize rich, configurable speech in real time and entirely on device, powered by AFM 3 Core Advanced, Apple's most powerful on-device foundation model. This work presents the memory-efficient audio synthesis architecture behind that capability: a detokenizer that converts the semantic audio tokens emitted by the foundation model into high-fidelity audio within the tight compute and memory budget of the Apple Matrix Coprocessor (AMX). We convert semantic audio tokens to a residual vector quantization (RVQ) representation with a three-component design, a streaming encoder, a temporal decoder, and a depth decoder, that systematically decouples temporal and depth processing. A single reusable depth decoder with Diffusion Transformer (DiT)-style stage conditioning generates all RVQ levels autoregressively, replacing the dedicated per-level decoders of prior multi-decoder architectures, while causal sliding window attention with fixed-window key-value caching yields constant memory complexity independent of sequence length. Deployed on the AMX, the detokenizer sustains roughly 10 ms per generation step, about 16x faster than real time, with a peak runtime memory of only 21 MB and 329 MB of on-device assets, enabling continuous streaming synthesis of 20-320 seconds of audio. This constant, small footprint replaces the linear and quadratic memory scaling of conventional transformer- and GAN-based approaches. Ablation studies validate the key architectural components, and audio quality assessment confirms that the architecture maintains synthesis fidelity while achieving efficiency gains over existing methods. Operating at a 1-billion-parameter activation size within AFM 3 Core Advanced, it improves Mean Opinion Score by +0.28 overall (4.15 vs. 3.87) and by +0.42 on conversational speech (4.24 vs. 3.82) over the prior on-device text-to-speech system.

Do learned audio embeddings encode structure that nobody told them to encode? We probe four large pretrained audio models (AST, CLAP, BEATs-bio and BirdNET) with a downstream task none of them saw during training: recovering phylogenetic distance from species vocalizations. If the geometry of the embedding space tracks the tree of life, the representation is picking up something deeper than the labels the model was optimized for. We run Mantel tests across two independent radiations. In 32 marine mammal species (1,754 recordings from the Watkins Marine Mammal Sound Database) the foundation models recover strong phylogenetic signal within the 26 cetaceans (CLAP r=0.82, BEATs-bio r=0.82, AST r=0.74; all p<0.001), among the highest acoustic-phylogenetic correlations reported for any taxon. Hand-crafted MFCC features (105d) find nothing (r=0.040, p=0.338). The gap survives after PCA-projecting every embedding down to 105 dimensions, so it is not an artefact of representation size. It also survives a partial Mantel test controlling for dominant frequency (partial Mantel r=0.404, keeping 97% of the variance explained), so it is not just pitch in disguise. We repeat the analysis on 20 bird species using the Jetz et al. (2012) phylogeny, and this time add BirdNET, a classifier trained end-to-end on around 6,000 bird species. The general-purpose foundation models recover the signal again (AST r=0.55, CLAP r=0.52). The unexpected result is that neither BirdNET nor the bioacoustic BEATs-bio beat them (r around 0.32 to 0.36). Matching the training domain to the target taxon does not, by itself, help. Pretrained audio embeddings carry evolutionary information across two independent radiations, and domain-specific pretraining is not required for it to emerge.

We present a self-supervised acoustic eavesdropping attack that reconstructs typed text solely from keystroke sounds, without requiring labeled data for the target device. The proposed attack enables stealthy eavesdropping in two real-world scenarios-physical spaces (public and semi-public) and online meetings. Our method combines unsupervised acoustic clustering with Transformer-based language model inference and iterative self-training, enabling stable character inference under highly uncertain acoustic-to-character mappings. We demonstrate that the proposed method achieves over 99% reconstruction accuracy with only 100-150 observed keystrokes under a close-proximity recording setup using a smartphone placed near the target device, significantly outperforming prior unsupervised baselines in low-data regimes. We further evaluate robustness across multiple laptop platforms and in realistic acquisition channels, including distance recording from approximately 3 meters away on the same desk, through-the-wall eavesdropping with a contact microphone, and background keyboard noise in online conferencing systems. Across these scenarios, the proposed method achieves high reconstruction accuracy (often exceeding 90%) with approximately 150-250 observed keystrokes. These results indicate that accurate text reconstruction from keystroke sounds is feasible in practice under an audio-only setting, even with limited observed keystrokes and without requiring device-specific labeled data, highlighting a realistic and previously underestimated privacy risk.

Audio deepfake detectors are trained to distinguish genuine speech from synthetic speech and often perform well on standard benchmarks. Yet the same detector that achieves less than 1% error on one dataset can see its error rate increase twentyfold when evaluated on a different dataset. We argue that one contributing factor is speaker-identity reliance: standard training corpora correlate speaker identity with the genuine/synthetic label, allowing detectors to partially rely on speaker-related cues rather than synthesis artifacts alone. We propose the Identity Sensitivity Score (ISS), a per-utterance diagnostic that quantifies how much a detector's output changes across different speaker identity contexts. ISS requires no ground-truth labels at inference time and can be computed from the detector score and a pool of reference speaker examples. Across two detectors and two datasets, incorrectly classified utterances have ISS scores 29 to 52 times higher than correctly classified utterances, and ISS alone predicts misclassification with area-under-curve (AUC) up to 0.954. To test whether ISS actually captures identity-sensitive behavior rather than serving only as a proxy for prediction confidence, we apply voice conversion to 500 utterances and measure the resulting detector-score shift. Utterances flagged as identity-sensitive by ISS respond 19 to 30 times more strongly to this manipulation than utterances flagged as stable. These results position ISS as a practical inference-time diagnostic for speaker-dependent failure analysis in audio deepfake detection.

While large audio-language models have achieved remarkable progress in auditory perception, they still lag behind text-based large language models in deep logical reasoning, primarily due to the scarcity of high-quality audio reasoning data. To bridge this gap, we propose X$^3$-OPD, a cross-modal on-policy distillation framework that transfers reasoning capabilities from a powerful text teacher to an audio-language student. During training, the student generates reasoning trajectories conditioned on its own acoustic perception, while the teacher provides token-level guidance using matched textual inputs and verified answers. We further construct a three-tier symmetric corpus covering textual reasoning rendered into speech, audio-event reasoning grounded in complex acoustic scenes, and spoken-dialogue reasoning involving paralinguistic cues. This design extends cross-modal distillation beyond textually recoverable content to reasoning grounded in non-linguistic events, prosody, and conversational context. Experiments on MMSU, MMAU, BIG Bench Audio, and MMAR demonstrate that X$^3$-OPD substantially improves audio-grounded reasoning and chain-of-thought quality while largely preserving the model's existing capabilities under domain shift.

Neural audio codecs are challenging transformations for audio watermarking because they re-encode, quantize, and resynthesize speech. This paper investigates continuous latent-space watermarking for codec robustness. Instead of adding a watermark only to the waveform or spectrogram, we embed a 32-bit message into the continuous latent representation of a codec-like speech autoencoder. The pipeline uses a SEANet-style encoder-decoder, a Conformer-based message embedder, RVQ-guided latent decomposition, and a latent-domain detector trained under signal-processing and neural-codec transformations. Rather than proposing a final universal watermarking baseline, we characterize the trade-offs that appear when the watermark carrier is moved before neural decoding. On 48 kHz speech, EnCodec-aware training improves EnCodec-24k bit accuracy from 78.8% to 95.6% and 97.1%, while PESQ decreases from 3.727 to 3.514 and 3.427.

Advances in AI-based voice conversion have enabled a wide range of media applications, including films, audiobooks, and games. However, most research and public benchmarks still focus on natural human speech, leaving designed vocalizations, such as monster growls and robotic voices, underexplored, partly due to the lack of publicly available resources. To address this gap, we introduce the Designed Vocalizations Dataset, constructed by curating diverse raw vocal sources, including speech and animal vocalizations, and applying professional vocal effects processing to produce corresponding effect modified variants. We further provide a standardized test set with explicit seen/unseen splits over source timbre groups and preset styles to assess generalization under controlled conditions. Finally, we report baseline benchmark results to support reproducible evaluation and future research. The dataset and demo samples are available at https://ncai-official.github.io/speech/publications/designed-vocalizations-dataset/.

Large Audio Language models (LALMs) have made rapid progress on acoustic understanding, yet they still struggle with fine-grained audio reasoning (e.g., recognizing event order, repetitions and duration). Existing post-training methods heavily rely on expensive external labels or provide only coarse semantic signals. To bridge this gap, we introduce Audio-Zero, the first label-free self-evolution framework in the field of LALMs that improves fine-grained auditory perception and reasoning. Audio-Zero constructs an auditory self-play game from unlabeled audio contrast pairs: most players hear a reference audio, while one odd listener hears a subtle variant. The model first generates clues describing what it hears and then identifies the odd listener by reasoning over inconsistencies among clues. Since the odd listener is known by construction, the game provides verifiable rewards without any annotated answers. Experiments with Qwen2-Audio-7B-Instruct and Qwen2.5-Omni-7B on TREA, MMAU Test-mini and MMAR show that Audio-Zero improves fine-grained audio reasoning while preserving broad audio understanding. Evolutionary and diagnostic analyses further reveal that increasingly fine-grained auditory descriptions emerge naturally from game pressure.

Passive acoustic monitoring of tropical biodiversity is bottlenecked by the storage and battery cost of continuously recording soundscapes in which bird vocalisations typically occupy less than 10% of the audio. Autonomous recording units built on low-power microcontrollers (typically ARM Cortex-M with $\leq$256 kB of RAM) address this by triggering only on likely-positive segments, but the on-device options are unsatisfying: coarse frequency-energy triggers such as Goertzel filters flood SD cards with false positives at $\sim$71% precision, whereas neural detectors developed for temperate single-species tasks are either too large to deploy or transfer poorly to species-rich tropical settings. We present DrongoNet, a family of three INT8 CNN detectors sized for this envelope and validated on a 50,000-clip, 1,677-species Southeast Asian tropical dataset (SEABAD). The headline model, DrongoNet-Micro (919 parameters, 6.26 kB, 0.9810 AUC, 98.3% mean recall at τ = 0.35), is a drop-in replacement for the Goertzel trigger used in commodity field recorders: at α = 0.10 tropical prevalence it captures 8 pp more bird vocalisations than Goertzel and extends a 32 GB card from $\sim$28 to $\sim$45 days of monitoring. DrongoNet-Nano (5.09 kB) bounds the ultra-low-flash extreme; DrongoNet-Edge (33.06 kB, 0.9991 AUC) targets Linux SBCs. On SEABAD, Micro matches a retrained TinyChirp CNN-Mel baseline within 0.1 pp AUC at 28$\times$ fewer parameters, confirming that the family is deployment-agnostic across mel-spectrogram bird corpora but requires per-environment retraining. Full INT8 quantisation costs $<$0.12% AUC across all three variants.

Dynamic Range Compression (DRC) is a widely used nonlinear audio effect whose parameters are often unknown, making blind estimation and inversion challenging. In this work, we formulate DRC parameter estimation as a black-box optimization problem in a perceptually motivated feature space. Given an observed signal and a reference representation, we estimate the parameters that minimize the distance between feature descriptors of the reconstructed and reference signals. Unlike gradient-based approaches, the proposed method does not require differentiability of the DRC model or the feature extraction pipeline, enabling the use of nonlinear and histogram-based descriptors. Experimental results demonstrate that the proposed method achieves competitive performance in blind parameter estimation and dry signal recovery, outperforming or matching state-of-the-art models in terms of reconstruction quality.

Almost every piece of recorded music you have ever heard was modified before it reached you; commercial releases rarely spring fully-formed from the mind of a musician. Despite the promise of music generation models for one-shot output, such fine-grained iterative refinement workflows are a complementary problem, and largely out of their reach. There is also a gap for musicians: while they can express what they want to hear, not all have the facility with studio production tools to implement the complex set of actions needed to realize these intuitions. We formalize this task as agentic post-production, wherein individual aspects of a song are targeted, refined, and combined into a final track. We argue the bottleneck is data: existing corpora do not reflect how realistic post-production chains map onto the vocabulary musicians and engineers actually use. We argue there is a language for modifying recorded music that is dense, consistent, and learnable. We introduce the Rule-based Instructions for Music Editing (RIME) framework, which generates realistic paired edit-instruction data from any baseline music dataset grounded in canonical methods, design patterns, and constraints derived from real production workflows. RIME leverages POEMS, a new toolkit that combines stem separation, mixing, and common studio effects for use by multimodal agents. We use RIME and POEMS to generate 3,000 pairs of edit instructions and ground truth audio, and use this data to evaluate existing multimodal LLMs as agents on this task, showing persistent challenges in current models' post-production capabilities. We also demonstrate RIME's ability to improve post-production agent performance via supervised fine-tuning. We see RIME as an early step toward iterative musical agents, collaborative systems that could transform music production much as interactive coding agents have reshaped software engineering.

Caption Studio is a transparency-first speech and audio intelligence platform that transforms spoken audio and video into structured, searchable content through automated transcription, speaker diarization, speech analytics, signal-level audio analysis, and subtitle generation. The system is built on a FastAPI backend with a real-time dashboard and adopts a three-layer architecture comprising (i) a transcription and diarization core based on Whisper-class automatic speech recognition and pyannote speaker diarization, (ii) an audio intelligence layer that extracts acoustic and linguistic features, including waveforms, spectrograms, pitch, speaking rate, silence, filler-word frequency, and sentiment, directly from the audio signal, and (iii) an integration layer that supports data export and downstream workflow integration. A principal contribution of this work is the transparency-first framework, in which every reported metric is explicitly identified as measured, derived, or unavailable, thereby improving the traceability, interpretability, and reliability of speech analytics. The paper presents the system architecture, benchmarking methodology, explainability and uncertainty framework, and key considerations for enterprise-scale deployment.

A single embedding space that covers text, images, video, and audio lets one index serve every query a user can pose. Embedding models built on vision-language backbones now lead text/image/video retrieval benchmarks but lack audio entirely, while audio-text retrieval is led by specialist systems that serve no other modality. We present the Fusion Embedding family, which adds audio to a frozen vision-language embedding base whose parameters are never updated: generation 1 (fusion-embedding-1) trains only a 16.4M-parameter connector between a frozen audio tower and the frozen base, and generation 2 (fusion-embedding-2) adds modality-gated deep adapters (44.2M parameters) whose branch never executes on text, image, or video inputs: their outputs are bit-for-bit those of the released base, verified after every training run. Because the base already binds text, images, and video, aligning audio to text alone makes audio-image retrieval emerge, with zero paired audio-visual training data. Alongside the recipe we map its design space with controlled negative results (rewriting training captions with an LLM, substituting a leaderboard-stronger audio tower, and widening the connector each reduce retrieval) and with training-protocol findings that we expect to transfer to any frozen decoder-LM embedding backbone. Both generations train in hours on a single GPU. Weights, code, and the evaluation harness are openly released.

This paper explores a way to verify that audio has not been maliciously tampered in a specific context: short viral videos taken from news recordings. Rather than trying to detect artifacts of tampering (internal inconsistency), we focus on positively verifying a query against a trusted source such as a news recording (external consistency). We propose a method for cross verifying a short audio query against a reference recording from which it was taken. Our approach is to define two hypotheses (non-tampered vs tampered), calculate the most likely alignment between query and reference for each hypothesis, and then perform a likelihood ratio test on the two alignments. We show that this method is fast to compute, much more robust than using MFCC features with Euclidean distance, and has the key benefit of explainability. Our cross verification approach provides an alternative perspective and complementary tool to existing tampering detection methods.

As large audio language models (LALMs) advance, robust evaluation frameworks have become essential. In this context, Spanish speech understanding under realistic acoustic conditions has received particularly little attention. We introduce ESCUCHA, the first Spanish speech understanding benchmark designed to evaluate LALMs across heterogeneous acoustic conditions and reasoning abilities. ESCUCHA comprises 1,000 human-curated questions paired with audio, totaling 162.9 hours sourced directly ``from the wild'' rather than drawn from existing datasets, with durations ranging from a few seconds to over 80 minutes. The benchmark emphasizes reasoning, spanning 9 perceptual and 10 reasoning categories, and it captures linguistic diversity through multiple Spanish accents and non-normative speech. ESCUCHA further includes multi-audio questions, spoken questions, and audio instructions, and it flags which questions support open-ended evaluation. Benchmarking several state-of-the-art multimodal and speech models reveals substantial performance gaps relative to trained humans.

Zero-shot text-guided editing of real-world music recordings requires balancing semantic modification with faithful preservation of the original musical structure. Although recent diffusion transformers trained with rectified flow have achieved remarkable success in text-to-music generation, extending them to edit existing recordings remains challenging because editing requires accurate deterministic inversion, reliable structural preservation, and numerically stable integration throughout the inversion and generation processes. We present FlowSonic, a zero-shot music editing framework built upon a pretrained diffusion transformer trained with rectified flow. FlowSonic first deterministically inverts a real-world recording into the latent space and preserves its musical structure during editing by reusing cross-attention representations extracted during inversion. To improve the numerical reliability of inversion-based editing, we introduce a high-order ODE solver and systematically investigate how different numerical integration schemes influence trajectory stability, structural preservation, and semantic controllability. Comprehensive experiments on timbre-transfer and genre-modification tasks demonstrate that FlowSonic consistently outperforms existing music editing methods across semantic alignment, harmonic preservation, structural consistency, and perceptual audio quality. We further provide geometric and empirical analyses showing how the proposed numerical integration strategy improves latent trajectory stability and leads to more reliable music editing.

This paper presents RealDESED, a real-world domestic sound event detection (SED) benchmark comprising 5,710 audio recordings collected by 652 participants in their homes. Each recording is between 15 and 35 seconds long and contains temporally precise annotations for 15 common domestic sound classes. In contrast to existing SED datasets, which typically rely on simulated soundscapes or broad web-crawled audio, RealDESED consists exclusively of recordings captured in natural domestic environments, reflecting realistic variability in recording devices, device placement, acoustic conditions, background sounds, and naturally occurring event co-occurrences. A distinguishing characteristic of the dataset is its multi-annotator labeling scheme, where each recording is independently annotated by multiple annotators, while the validation and test sets undergo an additional review process to ensure high annotation quality and reliable benchmarking. Furthermore, the dataset provides rich metadata, including recording device, device placement, environment labels, and textual scene descriptions. We establish a strong transformer-based baseline and investigate annotation aggregation strategies, post-processing methods, long-form inference, and the impact of recording metadata on model performance. Our baseline achieves a macro-averaged PSDS1 score of 0.731 on the test set. We believe RealDESED provides a valuable benchmark for developing and evaluating robust SED systems under realistic domestic conditions, helping to bridge the gap between current research benchmarks and real-world deployment.

We present Audio-Visual Flamingo (AV-Flamingo), a fully open state-of-the-art audio-visual large language model (AV-LLM) for joint understanding and reasoning over audio, images, and long-form videos. Unlike prior AV-LLMs that primarily focus on short clips, AV-Flamingo is designed for understanding and reasoning over long and complex real-world (audio-visual) videos. To support this, we make three key contributions: (i) Audio-Visual-Skills, a large-scale collection of real-world videos with ~7M caption and question-answer training instances designed to emphasize temporal, compositional, and cross-modal audio-visual reasoning; (ii) a novel three-stage curriculum that progressively trains the model from short-range perception to long-horizon multi-event reasoning; and (iii) Temporal Audio-Visual Interleaved Chain-of-Thought, a reasoning framework that explicitly grounds intermediate reasoning steps to timestamps in long audio-visual streams, improving temporal alignment and interpretability. Extensive experiments across 15+ audio-visual, omni-modal, audio, and vision benchmarks show that AV-Flamingo outperforms similarly sized open models by clear margins and remains highly competitive with, and in some cases surpasses, much larger open-weight and closed models, particularly on long and complex real-world audio-visual understanding and reasoning tasks. Beyond benchmark performance, AV-Flamingo exhibits strong real-world utility and transfers well to unseen tasks, highlighting its robustness and generalization ability.

Recent advances in text-to-speech and voice cloning make high-quality spoofing inexpensive and scalable, threatening voice authentication systems, especially automatic speaker verification (ASV). Existing defenses mainly address this threat through binary countermeasures (CMs) for deepfake detection or spoofing-aware speaker verification (SASV), where current systems are dominated by modular ASV-CM fusion and cascaded pipelines. Although large audio language models (LALMs) have shown promise on related audio tasks, including CM and ASV, their use for SASV remains unexplored, despite their capacity to produce natural-language rationales for auditing and robustness beyond discriminative predictions. This work systematically evaluates LALMs for SASV against conventional pipelines under zero-shot prompting, supervised adaptation, reasoning-oriented training, and reinforcement-learning-based optimization. Our results show that pretrained LALMs are near chance in the zero-shot setting, confirming that they are not natively suited to SASV, but that task-specific adaptation closes this gap. We further find that competitive SASV performance can be achieved through several distinct routes. These findings position LALMs as a promising and auditable foundation for unified SASV, while clarifying where conventional cascade systems still lead.

Rich internal representations of musical structure are essential for music understanding tasks such as machine-assisted music co-writing, yet self-supervised approaches for symbolic music representation remain underexplored, particularly those that encode the hierarchical multiscale nature of musical structures. We present MIDI-RAE-JEPA, combining a pitch- and time-shift equivariance objective with LeJEPA and a Swin Transformer V2 encoder to learn such hierarchical representations of symbolic music encoded as piano roll images. The time-shift equivariance objective encourages the model to internalize temporal musical relationships. The encoder is trained purely on self-supervised objectives -- including a masked embedding predictor (MEP) -- with collapse prevented via SIGReg. A separate decoder trained on the frozen encoder embeddings achieves reconstruction F1 of 0.995, and a flow matching generative model conditioned on those embeddings produces generations that closely match the pitch register and rhythmic density of the conditioning excerpt, while mismatched conditioning yields unrelated but musically plausible output. Learned representations outperform a Haar scattering transform baseline on a downstream emotion classification task, and embedding distances increase monotonically with pitch and time shift magnitude, confirming measurable equivariance. These results suggest that equivariance-based SSL objectives, combined with sufficient fine-level encoder capacity, provide a viable path toward semantically rich, generatively useful representations of symbolic music.

Bioacoustic foundation models rely on large-scale citizen science platforms like Xeno-Canto for geographically and ecologically diverse data. Recent work has shown that supervision alone can produce SotA species detection models when trained on this large-scale data -- however, there remains unutilized potential in the form of recording metadata readily available within these community-driven data hubs. In this work, we explore the use of metadata -- such as location and time -- as auxiliary supervision signals, allowing the model to leverage species-metadata correlations in its learned representation. Auxiliary metadata losses provide additional information beyond vocalizations alone that can encourage a richer, more robust representation that generalizes better to species distribution and acoustic domain shifts -- important challenges for deployment in real-world passive acoustic monitoring (PAM) settings. We introduce MetaPerch, a new foundation model that achieves strong species identification performance across multiple challenging domains and present an extensive empirical study of the effects of 9 diverse metadata sources on 17 bioacoustic datasets.

Bioacoustic call-type classification relies on costly expert annotation. Active learning can reduce this burden by selecting a small batch of segments for expert annotation and using the labeled segments for training the classifier. The setting is hard: the target calls are extremely sparse and the call-type distribution is long-tailed, so a tight budget must be spent on the few rare, informative segments. We propose BADGE-Greedy-DPP, a deterministic batch selector that greedily adds the segment whose BADGE gradient embedding most enlarges the volume spanned by the batch; because this log-volume objective is submodular, the greedy rule guarantees a batch value at least a (1-1/e) fraction of the optimum of this objective, a guarantee not provided by BADGE's existing k-means++ and MCMC DPP sampling heuristics. There is also a temporal granularity mismatch in the task. The acquisition function scores whole segments, yet the informative frames inside them are few. Uniform averaging therefore washes them out. We show that the BADGE construction naturally addresses this mismatch when applied frame-wise, as prediction residuals weight the aggregated pseudo-gradient, so confidently predicted no-call frames contribute little while a single uncertain rare-call frame can still set the segment's direction. Across 10 runs on a sparse, imbalanced hyena call-type dataset, BADGE-Greedy-DPP achieves the best overall and rare-call-type performance among all compared query strategies, including MFFT, the strongest non-BADGE baseline, and the two vanilla BADGE traversals.

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.

Recent text-to-audio models generate high-quality audio, but often fail to follow instructions involving multiple sound events and temporal order. This gap arises because existing evaluation and training signals mainly emphasize global similarity or perceptual quality, with limited supervision on instruction-level correctness. We propose an instruction-level framework that uses audio-aware large language models (ALLMs) as fine-grained judges to verify target event presence and temporal relations in generated audio. After validating ALLM judgments on benchmarks and through human verification, we use their feedback to construct preference pairs for direct preference optimization. We further introduce S3Bench, a narrative benchmark for evaluating multi-event temporal instruction following. Experiments show that our method improves event completeness, temporal ordering, and joint instruction-following accuracy across existing benchmarks and S3Bench, while maintaining audio quality.

A field can reformulate its computations freely exactly where its demand is stated independently of any incumbent implementation, and finds itself unable to when the incumbent's own output has quietly become the specification. This note offers that observation as a lens on computational reformulation for modern accelerators, where posing a problem in a hardware-friendly form can yield large speed and energy gains, but only if a replacement can be judged at all. Building on the test-oracle problem (Weyuker; Barr et al.), on requirements engineering's notion of implementation bias (Zave and Jackson), and on the roofline performance model, it names the pathology "baseline capture" -- the moment an incumbent stops being evidence that a demand can be met and becomes the definition of meeting it -- and separates two questions that are easily confused: whether a reformulation can be judged (which turns on the existence of an incumbent-independent demand) and whether its discovery can be automated (which turns additionally on the cost of evaluating that demand). Short cases -- shortest-path routing, learnable audio frontends, ZIP-215 for Ed25519 signature validation, CESM-ECT for climate models, and a single-GEMM audio frontend -- illustrate the pattern and the move of "buying a verifier": making a demand explicit, operational, and independent of the incumbent. No component is claimed novel in isolation; the contribution is the synthesis and the single question it makes easy to ask of any reformulation result -- does its acceptance test mention the incumbent's output?

Music recordings and live streams are often affected by noise, reverberation, spectral imbalances, or artifacts that degrade listening quality. While speech enhancement has matured into a well-defined research area, music enhancement is less established because musical signals combine overlapping sources, wide bandwidths, strong dynamics, and intentional production effects. We study real-time music enhancement under strict causal and low-latency constraints. We formulate the task around recovery of the intended produced mix from acoustic and production-oriented degradations, adapt compact causal networks to music, and compare speech-derived real-time baselines, an external music-denoising model, an offline restoration reference, and a music-specific MusicFilterNet-MS variant. On the tested hardware, all causal models run faster than real time, but improvements depend strongly on the dataset, degradation type, and metric family; under several objective criteria, indiscriminate enhancement can worsen the degraded input. The main contribution is therefore a benchmark and an analysis rather than a universal best model: real-time music enhancement is feasible, but robust improvement requires degradation-aware modeling, stereo-aware processing, identity-preserving correction, and evaluation beyond a single objective score.

Neural audio codecs were originally developed for high-fidelity compression; however, their latent token representations and expressive decoders also constitute a powerful substrate for controllable audio transformation. This work introduces Neural Morphing, a training-free token-domain audio effect that selects residual-vector-quantized (RVQ) token grains from a user palette and decodes the edited stream through a pretrained codec. The method combines an RVQ-group transfer policy that separates coarse, middle, and fine codebook groups with a continuity-constrained sequence matcher that replaces independent greedy selection with bounded beam search. The intended output is a controlled hybrid: the source preserves rhythmic organization while the palette contributes timbral color and residual detail. We focus on the implementation and realtime behavior of a deployable VST3/AU system, including chunked rendering, palette-size scaling, and backend health checks.

Since the paradigm centered on convolutional neural networks and recurrent architectures was established in 2020, the fundamental backbone networks for audio-visual navigation have undergone no essential changes for more than five years, making them inadequate to support efficient representation of dynamic multimodal sequences. This paper proposes Samba(A Hybrid Mamba for Audio-Visual Navigation). It uses the adaptive selection-enabled Mamba State Encoder (M-SE) to replace conventional GRUs for temporal aggregation, and constructs an Audio Mamba Encoder (AME) to remedy the limitations of convolutional operators in capturing global time-frequency dependencies in spectrograms. Experiments demonstrate that Samba exhibits exceptional generalization performance when facing unheard sound sources and unseen scenes. On the Matterport3D dataset, it improves the navigation success rate (SR) by 11.3% compared with existing state-of-the-art models, and the performance gain is even more pronounced on the Replica dataset, which features finer scene structures. Such modernized architectural reconstruction unlocks stronger embodied representation capabilities at a lower computational cost, thereby providing a highly robust technical pathway for paradigm evolution in the field of audio-visual navigation.

The rapid advancement of synthetic speech generation methods has made audio deepfake detection a critical challenge in multimedia forensics. While recent approaches achieve high detection accuracy, they typically rely on black-box architectures that offer limited interpretability and high computational complexity. In this paper, we propose an explainable-by-design audio deepfake detection framework based on Wiener-Hopf linear prediction, processed by a lightweight 2D Convolutional Neural Network (CNN). This design enables a direct and transparent connection between classification outcomes and the acoustic properties of the signal. Experimental results on benchmark datasets demonstrate competitive detection performance while maintaining significantly lower computational complexity compared to state-of-the-art solutions. The interpretability analysis using Grad-CAM reveals that the classifier focuses on low-order predictor coefficients and on silence and transitional regions, suggesting that the Wiener-Hopf predictor captures reverberation characteristics and subtle statistical inconsistencies in synthetic speech. Finally, robustness experiments show that fine-tuning effectively recovers detection performance under common post-processing degradations, including additive noise, MP3 compression, and telephone filtering.

Large audio-language models (LALMs) often underperform on fine-grained, non-semantic attributes of speech, such as a speaker's emotion, despite strong performance on speech content. Improving this without the cost of retraining calls for an effective inference-time intervention, yet most existing methods intervene only after the audio encoder and operate at a relatively coarse granularity. The encoder itself, where acoustic information is first extracted from the waveform, remains largely unexplored, especially at the level of individual neurons. We introduce IAAN, Identifying and Amplifying Acoustic Neurons, a training-free and label-free method that scores each feed-forward neuron in the audio encoder by contrasting its activation on the real waveform with that on a noise reference lacking the real audio's acoustic information. IAAN then amplifies a small set of the highest-scoring neurons at inference. Across ten non-semantic speech attributes, IAAN improves average accuracy by 25.7 points on Audio-Flamingo-3, 21.4 on Qwen2.5-Omni, and 9.7 on Kimi-Audio. It also improves a model already explicitly fine-tuned to prioritize acoustic evidence. In controlled comparisons, both the encoder locus and neuron-level selectivity prove necessary for this gain. Intervening after the encoder, at the decoding side or inside the language model, yields little to no improvement, or even deteriorates accuracy. The improvement also depends on which specific neurons are amplified, not merely on their number, confirming that IAAN's acoustic score succeeds in identifying the neurons that matter. These results show that a small, precisely targeted intervention inside the audio encoder is an effective and largely untapped way to strengthen the acoustic understanding of LALMs, opening a new direction for inference-time methods that improve acoustic perception through neuron-level access to the encoder.

Music creation is fundamentally a process of revision. Yet symbolic music generation remains dominated by paradigms that produce complete sequences from scratch, with limited support for selective modification. Edit-based methods have proven effective for text transformation tasks, but remain largely unexplored for symbolic music. We trace this absence to the representational level: conventional event-based music encodings lack the structural properties required by explicit music editing. In contrast, the BEAT encoding, a beat-grid-anchored representation originally designed for autoregressive generation, possesses structural properties amenable to editing. We propose BeatEdit, the first framework for symbolic music generation based on explicit edit operations, recasting generation as producing new content by editing a draft rather than synthesizing from scratch. BeatEdit comprises three complementary mechanisms along an axis of increasing edit density: per-token sequence tagging for error correction, iterative refinement for accompaniment editing, and tag-then-fill for segment completion. All these mechanisms share a single encoding and pre-trained backbone, achieving higher precision and perceptual quality than autoregressive and diffusion methods across all three tasks, while remaining efficient, with single-pass inference completing in under 100 ms. Cross-encoding evaluation further reveals that encoding design substantially influences editing effectiveness, with notable encoding-method interaction effects. Code is available at https://github.com/Haoyu-Gu/BeatEdit-code

Dance-to-music generation is a promising task for applications such as choreography support and automatic accompaniment, where temporal coordination between body movement and sound is essential. In particular, using human joint positions as the motion representation is attractive because they explicitly capture body dynamics while being lightweight, privacy-preserving, and easy to integrate with motion capture and pose-estimation pipelines. A central challenge in this setting, however, is the scarcity of high-quality paired dance-music data, since collecting accurately synchronized pairs is costly and often constrained by copyright and performance rights. This makes it difficult to train end-to-end models solely from paired data. To address this issue, we propose a dance-conditioned music generation framework that efficiently exploits both unpaired and paired data. Our method combines pretrained unimodal encoders for motion and music, beat-guided contrastive pretraining to align their feature spaces, and a ControlNet-style conditioning module on top of a pretrained text-to-audio diffusion model. Experiments on AIST++ demonstrate that the proposed techniques improve both dance-music alignment and audio quality, as confirmed by quantitative and qualitative evaluations. Compared to a state-of-the-art method, our approach achieves superior dance alignment performance and competitive audio quality. Code is available at https://github.com/kmraven/AudioLDM-ControlNet .

While recent few-step sampling text-to-audio generation models like MeanAudio substantially accelerate generation by modeling average velocities, their strict one-step generation quality still lags significantly behind multi-step counterparts. We propose FdAudio to bridge this gap. Unlike MeanAudio, which relies solely on regression against target velocity fields, our post-training approach optimizes the final one-step distribution directly across pre-trained embedding spaces via a multi-representation Fréchet-distance (FD) loss. Crucially, to prevent the multi-step degradation that naive post-training with FD-loss causes, we introduce a MeanFlow consistency objective as a structural anchor. Results demonstrate that FdAudio establishes state-of-the-art one-step T2A generation quality among few-step systems, yielding an 11.4% reduction in FD score and a 28.8% improvement in FAD score relative to the baseline MeanAudio framework. Notably, we solve FD post-training's naive multi-step degradation issue by proposing the MeanFlow anchor, enabling a 25-step sampling path to maintain high-fidelity audio synthesis that matches or surpasses strong multi-step models at a fraction of their computational latency.

Temporal grounding in long recordings remains challenging for audio-conditioned LLMs. We present a time-aware audio LLM that answers questions with explicit timestamps over up to 120 minutes of input. Our approach interleaves periodic time markers with continuous audio tokens using large-scale synthetic supervision from a cascaded pipeline. Our model achieves strong temporal-grounding accuracy on short and long benchmarks and supports time-anchored fragment descriptions and summaries. Extensive ablations examine how time representation, marker frequency, tokenization, and duration-mixture design affect accuracy and computational cost. We release model weights and datasets to support further research on time-aware audio understanding, available at https://huggingface.co/ai-sage/GigaChat3.1-Audio-10B-A1.8B.

Recent advances in large-scale multimodal models have drivenremarkable progress in vision-language tasks; however, comprehensiveomni-modal understanding remains under-explored, largely due to thescarcity of datasets with rich, explicitly aligned auditory cues. To bridgethis gap, we present AVDC (Audio-Visual Decoupled Captions), a large-scaledataset designed to disentangle visual and auditory semantics. Specifi-cally, we propose an automated pipeline that leverages off-the-shelf mod-els to annotate videos with tripartite captions: visual-only (V), audio-only (A), and joint audio-visual (AV). This decoupled structure explic-itly captures both modality-specific nuances and complex cross-modalinteractions. Building upon this, we introduce AVDC-QA-CoT, a Chain-of-Thought augmented question-answering dataset to foster audio-visualreasoning. To fully exploit these resources, we employ a two-stage train-ing paradigm: omni-modal caption generation pre-training on AVDC, fol-lowed by instruction tuning on AVDC-QA-CoT. Extensive experiments acrossdiverse downstream tasks, spanning video captioning, audio-centric anal-ysis, and omni-modal benchmarks, demonstrate consistent and signifi-cant performance gains, showing the efficacy of our proposed datasetsand training strategy in advancing omni-modal perception. Code anddataset are related on https://radiant0726.github.io/AVDC-web/.

Generative streaming models for Target Speaker Extraction (TSE) commonly exhibit a quality--intelligibility trade-off, wherein naive optimization for perceptual audio quality tends to degrade speech intelligibility, and conversely. We reveal that this trade-off arises not from the constraints of streaming architectures, but from an inappropriate choice of optimization anchor. Directly optimizing against audio quality metrics induces catastrophic reward hacking, where content critical to pronunciation and intelligibility is systematically erased to maximize a proxy score. To break this bottleneck, we propose two complementary improvements: an enlarged Conformer convolution kernel for richer local spectro-temporal modeling, and WavLM-anchored Direct Preference Optimization (DPO) fine-tuning strategy. DPO preference pairs are ranked by WavLM cosine similarity, a deep acoustic feature encoding both phonetic structure and speaker identity, providing an optimization anchor that resists hacking. Under a 560 ms streaming chunk size, the proposed method achieves a 10.9% relative intelligibility improvement (word error rate: 0.138 to 0.123), with marginal simultaneous gains in audio quality and speaker similarity.

Understanding music requires understanding localized relationships across data modalities, e.g., how time in performance audio maps onto position in a score image. Yet supervision for such local correspondences is difficult to obtain-in practice, we often only have access to coarser global supervision like paired segments of audio and images. To address this gap, we propose FuSiLi (Fused Sinkhorn-Localized Similarity), a similarity score for multimodal contrastive learning operating directly on local image patch and audio frame features via Sinkhorn-based soft alignment. We show that FuSiLi (i) effectively learns local relationships, (ii) requires only global supervision, and (iii) retains the global alignment capabilities of conventional contrastive approaches. We fine-tune pretrained CLIP and CLAP encoders on pairs of raw sheet music images and audio using a hybrid contrastive objective combining FuSiLi with conventional global similarity. We evaluate on cross-modal retrieval and frame-level alignment tasks against a range of global and local baselines, showing that our approach outperforms them on local alignment while remaining competitive on retrieval.

Self-supervised learning for symbolic music has advanced largely through token-level pretraining, but such representations remain tied to tokenizer-specific sequences and often provide time-span-level embeddings only indirectly. In this paper, we propose ARIMA, a reconstruction-grounded latent predictive framework for symbolic music that learns compact window-based representations directly from data. ARIMA encodes each fixed-duration window into a continuous latent representation, trains a causal predictor with contrastive next-latent prediction, and grounds the encoder through structured reconstruction of music elements. This design preserves local musical details while modeling temporal progression across windows. We evaluate ARIMA on downstream tasks spanning various levels of music understanding. Results show that ARIMA is particularly efficient and effective on tasks involving harmonic, timing, and cross-performance retrieval, while remaining competitive with much larger baselines on other tasks. Ablations further show that next-latent prediction is essential for temporally integrated representations, and that structured reconstruction stabilizes latent learning without requiring explicit variance regularization. The code is at https://github.com/AndyWeasley2004/symbolic_music_wm.

Audio deepfake detection models determine whether speech is genuine or artificially generated, but high overall accuracy can mask substantial performance disparities across demographic groups. In this work, we investigate gender bias in audio deepfake detection using the ASVspoof5 dataset. We use ASVspoof5 under a controlled custom split designed to isolate gender-composition effects. We train attack-specific models on nine training sets with different gender compositions, ranging from female-only to male-only. We use a ResNet18 classifier with LogSpectrogram and WavLM-Base+ features, and we evaluated six post-hoc threshold calibration methods. Experimental results show that training data composition strongly predicts bias direction, with the underrepresented gender performing worse at test time. WavLM-Base+ features are shown to produce gender performance gaps 3.0 to 4.3 times larger than LogSpectrogram under identical training conditions, and balanced training is found to reduce LogSpectrogram bias but leave WavLM bias largely intact. Moreover, all six calibration strategies, including Oracle calibration with full test-set label access, leave the Equal Error Rate gap unchanged at 1.317 pp, confirming that threshold adjustment cannot correct underlying score distribution disparities. Overall, these findings suggest that gender fairness in audio deepfake detection must be addressed at training time, as post-hoc methods can only partially mitigate the resulting disparities

Representation alignment (REPA) has been investigated to accelerate diffusion training, but we observe that regularizing intermediate representations in diffusion Transformers (DiT) may implicitly entangle latents and limit generative capacity. To address this issue, we propose ReGen, a hierarchical multi-prompt representation generation framework that jointly estimates multiple vector fields for both representations and data within a single diffusion model. We further introduce generalized flow matching (GFM) to improve the generalization of conditional flow matching (CFM). We validate ReGen on single-stage waveform diffusion models including neural audio codec and Wave-VAE. ReGen significantly improves waveform generation quality from highly compressed latent representations at 12.5 Hz. We also present ReGenVoice, a latent diffusion model (LDM)-based text-to-speech model that achieves strong speech intelligibility (WER) and speaker similarity (SIM) with a small dataset. Moreover, operating the LDM at 6.25 Hz with rich semantic and acoustic latent representation enables efficient training and sampling, requiring only 1 day of training on 4 GPUs and fast inference with an RTF of 0.08. Audio samples are available at https://regenvoice.github.io/demo/.

We present Clean2FX, a study and demo of label-conditioned clean-to-effect transformation for electric guitar audio. Given a clean guitar input and a target effect label, the task is to synthesize the corresponding effected signal while preserving the musical content. Training and evaluation pairs are constructed from EGFxSet real, single tone recordings by assembling matched clean/effected chords, melodies, and mixed timelines. This allows for controlled comparison across effects. We evaluate four neural approaches under a common spectrogram-based transformation setting: two variational autoencoders and two U-Net models that differ in whether they operate on linear or log-magnitude representations. Performance is measured using linear-magnitude spectrogram MSE and Fréchet Audio Distance. The U-Net models outperform the variational autoencoder variants. Per-effect results show that distortion effects are most readily improved, whereas delay and reverb effects exhibit weaker FAD gains despite substantial spectral-error reductions. A conditioning-sensitivity diagnostic provides evidence that the best model responds to target labels rather than collapsing to a single transformation. Our demo website compares two models applied on real-world guitar performances outside training and validation data, providing audio and spectrogram examples of the practical clean-to-effect behavior.

Multimodal Large Language Models (LLMs) have remarkable semantic audio understanding, yet they remain "spatially agnostic" due to their reliance on mono-channel audio representations. Currently, spatial audio perception methods mainly focus on complex room simulations and custom-trained, geometry-aware stereo encoders, which limits their accessibility and generalizability. In this paper, we introduce the Dual-BEATs architecture, in which the left and right audio channels are routed independently through two identical semantic encoders as an alternative to specialized spatial modules. To circumvent the architectural bottleneck where internal normalization otherwise erases the inter-channel variance of stereo audio, we inject a static, uncorrelated dithering noise floor prior to encoding. This dithering intervention establishes a macro-variance floor that "smuggles" spatial geometry across the normalization layers. Evaluated on a ternary directional classification task (Left, Center, Right), we demonstrate that dithered models achieve exceptional spatial resolution--reaching up to 97.2% localization accuracy even on subtle 0.5 panning amplitudes--and demonstrates robust, zero-shot generalization to entirely unseen spatial configurations. Our results suggest that with the appropriate acoustic regularization, standard multimodal models are natively capable of generalized stereo audio understanding.

With the growing adoption of VLMs, DMs, LLMs, and AFMs, these multimodal foundation models can inadvertently encode sensitive, copyrighted, biased, or unsafe cross-modal associations that originate from their training data. Retraining after deletion requests or policy updates is often impractical, and targeted forgetting remains difficult because knowledge is distributed across shared representations. Multimodal unlearning addresses this challenge by enabling selective removal across modalities while retaining overall utility. This survey offers a unified, system-oriented view of multimodal unlearning across vision, language, audio, and video, grounded in recent advances, emerging applications, and open problems. Our taxonomy enables systematic comparison across model architectures and modalities, clarifying trade-offs among deletion strength, retention, efficiency, reversibility, and robustness. This survey highlights open problems and practical considerations to support future research and deployment of multimodal unlearning. We release a curated repository: https://smsnobin77.github.io/Awesome-Multimodal-Unlearning/

Video-to-audio (V2A) generation aims to synthesize realistic audio that is both semantically consistent with and temporally synchronized to a silent video. Despite recent progress, many methods still rely on multi-stage training, resulting in high computational costs and long runtimes, or transform visual input into text to leverage pretrained text-to-audio models, sacrificing fine-grained temporal cues. To overcome these limitations, we propose Flowley, an end-to-end, single-stage training architecture that produces soundtracks by combining visual features with textual prompts. Crucially, we introduce Progressive Soft-masked Cross-Attention, which embeds audio-visual synchronization directly within its attention mechanism, adding zero additional computational cost compared to standard attention layers. We further observe that existing V2A benchmarks lack sound-oriented descriptive captions, which can potentially degrade the quality of the synthesized audio. To remedy this, we propose SoundCap, a plug-and-play pipeline for creating detailed, sound-aware captions that guide the model. Remarkably, without integrating any pretrained audio-visual alignment modules, Flowley achieves state-of-the-art performance on VGGSound across multiple metrics. Moreover, by incorporating SoundCap, we further exceed the performance of the strongest existing close-sourced methods in terms of audio quality in the zero-shot setting.

Chamber music, as a highly precise multi-part interactive system, contains a logic of "role assignment and dynamic interaction" that provides an extremely valuable blueprint for exploring human-computer collaborative composition paradigms. Addressing the lack of role perception capabilities in existing deep music generation models during polyphonic interactions, this paper conducts an interdisciplinary analysis of Haydn's String Quartet in D Major, The Lark (Op. 64, No. 5). We propose a novel research path: "Classical Morphology Qualitative Analysis-Electroacoustic Quantitative Measurement-Machine Representation Reconstruction." The study first utilizes auditory analysis to dissect the counterpoint morphology of the leading voice and the underlying groove in the first movement. Subsequently, it introduces spectrum and dynamic feature analysis tools from a Digital Audio Workstation (DAW) to translate subjective auditory perception into objective, measurable physical parameters. Building on this, the paper introduces a fundamentally new approach to low-level computer feature extraction: completely abandoning the traditional mechanical quantization grid, introducing Event-based Timestamps to record the duration of micro-timing, and transforming acoustic features into an independent "Role-Aware Encoding" as an aesthetic heuristic mechanism (a phenomenological anchor). This study not only completes the logical loop spanning classical analysis, electronic music mapping, and AI symbolic generation but also establishes a profound theoretical foundation-from the perspectives of interactive aesthetics and media philosophy-for constructing human-computer collaborative music systems imbued with "social attributes" and "otherness awareness."

Automatically recognizing the sentiment, positive or negative, from speech is a challenging task, requiring both the analysis of vocal inflections and the interpretation of uttered words. Recent solutions rely on audio foundation models to solve the task, but it remains unclear if such models can take all aspects into account. To this end, we propose a multimodal solution that integrates audio and text information via cross-modal transformers, where text transcripts are automatically generated via an automatic speech recognition (ASR) tool. Moreover, we create multiple text modalities by automatically translating the transcripts into multiple languages via machine translation tools. Audio and multilingual text features are combined via a cascaded architecture comprising cross-modal transformer blocks that integrate modalities one by one. We further distill knowledge from the multimodal model, called teacher, into a unimodal (audio only) model, called student. We conduct experiments on a large-scale dataset, demonstrating that the automatically generated textual information can bring significant performance boosts in multimodal sentiment polarity classification. Our ablation study confirms that both automatic transcripts and automatic translations are helpful. Moreover, we show that the audio-only model can be enhanced via distillation, boosting performance without any computational overhead during inference. To reproduce the reported results, we publicly release our code at https://github.com/andreidurdun/cross-modal-audio-sentiment.

Audio intelligence involves understanding, reasoning about, and generating both audio and speech. In this work, we introduce Nemotron-Labs-Audex-30B-A3B (Audex), a unified audio-text LLM built on Nemotron-Cascade-2-30B-A3B, a strong text-only MoE LLM. Audex adopts a simple unified design with a single Transformer decoder: audio inputs are encoded and projected into the text embedding space, while text tokens and quantized audio output tokens are treated uniformly during generation. This architecture enables strong audio-text fusion, seamless multimodal generation, and compatibility with standard LLM training and inference infrastructure. For training, we meticulously curate audio-text datasets comprising 157.4B audio tokens and 320.5B text tokens. We apply multi-stage supervised training on these datasets, followed by text-only Cascade RL and multi-domain on-policy distillation. Audex delivers state-of-the-art audio understanding, speech recognition and translation, text-to-speech, audio generation, and speech-to-speech generation, while preserving very compelling reasoning, alignment, knowledge, long-context, and agentic capabilities of its text-only LLM backbone with marginal or no regression. We release the model checkpoints to facilitate open research.

Language understanding in the brain is context-dependent, varying across experimental stimuli and individuals, which makes it difficult to build computational models that generalize across both. This calls for a foundation model of language-evoked brain activity that can capture shared structure while adapting efficiently to new participants and inputs. We introduce RABBiT (Rapidly Adaptive BOLD foundation model via BraIn-Tuning), a compact audio-to-fMRI encoder designed for accurate zero- and few-shot prediction. A comprehensive evaluation on 324 participants across multiple unseen fMRI datasets shows that RABBiT enables accurate zero-shot prediction of fMRI responses to natural speech across auditory and language-selective regions, surpassing the SOTA foundation model for fMRI and predictions based on group averages. With as little as 10 minutes of participant-specific data, RABBiT further improves performance via parameter-efficient tuning, substantially outperforming per-participant linear models. RABBiT's performance is driven by two key innovations: (1) learned region-specific attention, and (2) a decomposition of brain responses into shared and subject-specific components, combined with a brain-tuned speech backbone. In addition to supporting strong predictive accuracy, the structured, region-specific representations that RABBiT learns enable interpretability. By eliminating the need for extensive per-participant data and model fitting, RABBiT enables scalable population-level analyses of language in the human brain. We make the code available at https://github.com/bridge-ai-neuro/rabbit.

Modern automated audio captioning systems pair a frozen audio encoder with a large language model (LLM) via a trainable projector, incurring the encoder's inference cost and bottlenecking the model through its fixed acoustic features. We present CARD, an encoder-free audio captioning model that removes the encoder at inference: a 13.2M projector feeds a frozen LLM with merged LoRA adapters, while the teacher used to train it is discarded. CARD distills a pretrained audio teacher (CLAP-HTSAT) into the model, but rather than injecting it into the LLM alone, it routes the teacher's representations across components: perceptual stages to the projector and semantic stages to the LLM. This placement improves CIDEr-D by +12.18 over an LLM-only distilled model on AudioCaps and by +5.21 on Clotho, reaching 55.4 against a 66.4 encoder-kept upper bound with no encoder at inference, showing that where a teacher's knowledge is placed matters as much as its presence.

With the proliferation of AI-generated content, sophisticated multimedia manipulation has raised critical concerns about malicious applications such as opinion manipulation and evidence fabrication, making Audio-Visual Temporal Forgery Localization (AV-TFL) an urgent research frontier. Existing TFL methods have progressed along two main paradigms: Transformer-based temporal modeling and channel-wise multimodal fusion. While these approaches capture temporal dependencies and cross-modal correlations, they process all frequency components indiscriminately, leading to overfitting on high-frequency noise and limited robustness under real-world data degradation. Through systematic frequency domain analysis, we find that forgery-discriminative patterns concentrate in the low/mid-frequency range (normalized frequency 0-0.15), while high-frequency components primarily introduce noise, removing them even improves detection performance by +1.4%. Based on this phenomenon, we propose UniSkip-Mamba, a frequency-aware State Space Model framework that incorporates Unified Multimodal Sequence Fusion to preserve cross-modal phase relationships, and Skip-Scanning Mamba Blocks that implement frequency-aware regularization through a novel Group-Scan-Merge mechanism, naturally biasing learning toward discriminative low/mid-frequency patterns (0-0.15) while maintaining representational completeness. We achieve state-of-the-art (SOTA) performance: 63.4% AP@0.95 on LAV-DF (+9.8% improvement) and 63.58% mAP on AV-Deepfake1M (+14.32% improvement), with 6x faster inference. Our frequency-domain analysis provides theoretical justification from a signal processing perspective for why skip-scanning inherently improves both accuracy and robustness.

The performance of acoustic machine learning systems is commonly evaluated using fully annotated test sets. In real-world deployments, however, exhaustively labeling large volumes of continuously collected audio data is often infeasible. Consequently, performance assessment typically relies on a small labeled subset of the available data, introducing a sampling bias that can severely distort evaluation metrics. This paper studies methods for compensating the bias in evaluation-labeled subsets under strict annotation-budget constraints. We study whether importance weighting techniques can mitigate this discrepancy by compensating for the selection bias. Specifically, we implement and compare three density-ratio estimation methods: kernel density estimation (KDE), logistic regression, and k-nearest neighbors (kNN), utilizing feature-space representations of the deployed audio. To emulate realistic deployment scenarios, the labeled subsets are generated using five distinct sampling strategies based on active learning techniques. Experiments conducted on an audio scene classification (ASC) benchmark demonstrate that importance weighting consistently yields more realistic accuracy estimates, significantly reducing the gap between subset-based metrics and the true evaluation performance.

Large Audio-Language Models (LALMs) reason fluently about sound yet struggle to localize precisely when events occur, while classical Sound Event Detection attains frame-level precision only over a closed label set. At the intersection of these paradigms lies the task of Open-Vocabulary Audio Event Grounding: predicting all time intervals of a target sound event described by an arbitrary natural language query. While this task is crucial for real-world audio understanding and LALM adaptation, it is bottlenecked by data scarcity. Few large-scale resources provide open-vocabulary onset/offset supervision, and manual temporal annotation is prohibitively expensive. To address this, we introduce Auto-AEG, a scalable pipeline that constructs such supervision by automatic data construction and model fine-tuning. It pairs programmatically synthesized clips, which carry exact ground-truth intervals for supervised cold-start, with multi-model pseudo-labels on real-world audio that supply the reward signal for reinforcement learning. Training with this pipeline yields promising performance gains on both the DESED SED benchmark and AEGBench, an independent difficulty-stratified benchmark we release. Our results show that automatically constructed data, coupled with interval-aware reward function design, is an effective data-side route to expanding the temporal localization capability of LALMs.

Audio-conditioned generators now produce synthetic sound effects from real recordings, so the real and synthetic versions of an event increasingly coexist in sound libraries and in the corpora used to train audio models -- yet no benchmark measures whether a representation can match a synthetic clip to the specific real recording it was generated from. I introduce Doppelganger, a benchmark for matching sound effects across the synthetic-real boundary, pairing 10,420 real clips across 34 everyday sound events each with an audio-conditioned synthetic twin, alongside a controlled 7-class corpus. Off-the-shelf audio encoders do not cross the boundary cleanly. Making the embedding ignore the boundary the standard way -- training it on sound-event labels -- works on familiar sounds but backfires on new ones, dropping below the untrained encoder. Training on the pairs instead -- a clip and its own synthetic twin -- generalizes. On sound events held out of training, it recovers the exact real source about 80% of the time (up from 61% untrained; chance 0.03%), whereas no objective meaningfully improves category-level recognition on those unseen events. The learned matching is specific to one generator -- it survives changes to that generator's settings but not a switch to a different generator, and collapses for the text-only generators tested. A human annotation baseline (49 listeners) lands well above chance but below the models on the same trials. Synthetic twins fool people into calling them real about 29% of the time, yet a generator-specific detector separates these audio-conditioned twins from real recordings perfectly.

Audio foundation models are widely adopted as general-purpose feature extractors, yet the internal structure of their learned representations remains insufficiently understood. In this work, we analyze CLAP audio embeddings through a probing framework, studying the encoding of three fundamental perceptual dimensions: reverberation (RT60), loudness (LUFS), and spectral content, measured via spectral centroid (SC) and relative pitch (RP). Probes of increasing complexity are trained to predict each attribute from frozen embeddings across five datasets spanning noise, speech, monophonic musical notes, and music mixtures. Our primary finding is that all of these attributes are reliably recoverable from the CLAP embedding space across the examined datasets. Within this global picture, two encoding regimes emerge: RT60, LUFS, and RP are approximately linearly encoded, while SC requires non-linear probes. Both regimes generalize across eight additional audio foundation models, with the notable exception that amplitude-invariant architectures discard loudness entirely by construction. The identified linear feature directions are geometrically consistent across datasets for RT60 and LUFS, while highly domain-specific for RP. Finally, we provide a qualitative demonstration of cross-modal consistency, showing that text embeddings of acoustic descriptors align geometrically with the identified RT60 feature direction.

Crossmodal correspondences between sound and taste are well established in psychology and neuroscience, but largely absent from content-based multimedia retrieval. We formalise taste-from-audio prediction as a content-based music information retrieval benchmark over a perceptually validated multi-source corpus, comparing ten frozen audio encoders from the four HEAR families under a shared multi-task regression head, with gated late-fusion as a configurable variant. In order to assess the effectiveness of the models, we compute absolute error and rank correlation. The strongest systems predict the five tastes within a macro RMSE of 0.134; on held-out real music their error is less than half a single rater's deviation from the consensus (RMSE 0.13 vs. 0.28), so the model tracks the group consensus more closely than an average human rater, and well below the previous state of the art baseline (0.219). On absolute error the encoders are statistically flat, with a single VGGish matching the best fusion, but gated late-fusion's advantage is confined to rank correlation (macro Pearson r 0.724 vs. 0.666). Operationalised as a content-based retrieval index, the predicted taste space ranks a 309-item pool far more faithfully than a CLAP-text baseline, which sits at chance; ridge probes and an audio-bandstop knockout read the strongest representations against documented sound-taste correspondences.

Narration is central to the audiobook listening experience, shaping how listeners engage with and understand the content. This work explores how narration qualities shape an audiobook's appeal, noting that their effects can vary by genre, title, and audience. We extract vocal and acoustic features (e.g., tone, pace, loudness) from LibriVox using pre-trained audio models and analyse their relationship with consumption data (specifically, view-rate) and their interplay with genre and title. Despite limited consumption data, we find that acoustic information alone has a robust association with appeal, even after accounting for title effects. We further validate these findings using more nuanced proprietary engagement metrics. To our knowledge, this is the first systematic computational study linking narration qualities, genre, title, and audiobook consumption, highlighting the potential of data-driven insights to improve audiobook personalisation and narrator casting.

We present GRAFT, a per-word pronunciation conditioning mechanism for text-to-speech neural codec language modeling. Existing systems reach high intelligibility and naturalness but inherit the ambiguity of text and mispronounce rare proper nouns, loanwords and technical terms. Even phoneme-conditioned models offer no direct acoustic handle for per-word pronunciation. GRAFT controls the pronunciation of a chosen word from a short spoken sample of it, encoded with the model's own speech tokenizer and bound to the word's position in the prompt. Voice conversion during training-data construction disentangles the hint speaker from the target speaker, so the hint may come from any voice while the output stays in the target voice. In a blind English listening study, human raters rank GRAFT first by a clear margin, judging its rendering of the difficult word closest to a reference recording of that word. On a five-language objective benchmark, GRAFT reduces target-word phoneme error rate by 22-39% over the identical text-only backbone and outperforms competitive open-source zero-shot systems, both phoneme- and text-conditioned, on target-word pronunciation, while preserving speaker similarity and naturalness.

While Large Multimodal Models excel in comprehension, high-throughput inference engines lack native support for multimodal generation. This is severe in Speech Language Models, where generating multi-layered audio tokens via decoupled AR+NAR or synchronous Multi-Token Prediction (MTP) with delay-pattern interleaving conflicts with standard single-stream loops. We present a vLLM-based inference pipeline for unified speech understanding and generation. We extend autoregressive decoding to natively execute delay-pattern de-interleaving and coordinated multi-stream sampling, integrating an on-GPU acoustic decoder for end-to-end waveform synthesis. Crucially, we overcome the shared intuition that Classifier-Free Guidance (CFG) halves throughput. By co-scheduling paired conditional and unconditional requests within a continuous batch, our CFG implementation sustains 80% of non-CFG throughput, absorbing dual-request and logit merging overheads. We open-source our framework.

This technical report describes our system for Task 1 of the DCASE 2026 Challenge, which aims to classify heterogeneous audio recordings according to the Broad Sound Taxonomy (BST). The task requires both accurate second-level prediction and consistency with the top-level taxonomy. Our system is built on CLAP-based audio-text representations and is improved along three strategies: expanding the training set with a filtered subset of BSD35k, enhancing acoustic modeling with feature-specific branches, and refining predictions using hierarchy-aware classifiers and KNN-based post-processing. Among the acoustic features considered, the log-STFT branch provides the strongest single-model performance. With KNN-based post-processing, our best single system achieves a hierarchical F1 score (Hier. F1) of 80.84% on the BSD10k-v1.2 set under the same evaluation protocol as the baseline. We further construct ensemble systems by combining models with complementary acoustic features and classification heads, achieving Hier. F1 scores of 81.25% and 81.18%, respectively.

Discrete tokens obtained from neural audio codecs (NACs) have been used as compact representations in audio generation and understanding models. In such token-based systems, token temporal resolution (TTR), defined as the time interval between adjacent token frames, is important because it controls the trade-off between representing rapid acoustic events and reducing token-sequence length. However, most NACs are trained at a single TTR and require separate training for each TTR. This paper proposes a mechanism that enables a single NAC to operate at multiple TTRs using sampling-frequency-independent convolutional layers. The mechanism regards TTR as the sampling period of the token sequence and generates TTR-dependent convolutional kernels from a shared parameter set, while adjusting the kernel size and stride for each TTR. We incorporate the mechanism into Descript Audio Codec, leaving the quantizer unchanged. Experiments on environmental sound reconstruction show that the proposed model outperforms a single-model baseline that switches TTR-specific layers for each TTR.