Researchers present AURA, a multimodal framework that lets users edit music through a back-and-forth conversation rather than one-off commands. Existing instruction-guided editors handle each request in isolation, which makes it hard to iteratively refine a track. AURA is designed to keep track of prior edits so a session can build progressively toward a desired result.
A new arXiv preprint proposes using predictive audio representations to spot and track hidden dynamic objects before they come into view. The authors argue that occluded traffic agents can appear too late for detection systems, and that audio cues offer an earlier warning signal. The work targets safety-critical settings such as autonomous driving.
A new arXiv paper introduces SpecAugment-Patch Merging, a method that masks input spectrograms at the patch level before positional embeddings are added, then merges patches to cut computation. The authors describe it as a simple approach to accelerate training of Audio Spectrogram Transformers. The work is categorized under machine learning research.
A new study examines how audio language models attribute speech to the correct speaker, finding accuracy of only 6 to 16 percent on a six-speaker task, below random guessing. The authors show that speaker tracking relies on attention heads in the model's text backbone, and that altering a subset of those heads shifts which speaker the model retrieves.
Researchers present MAST, a framework for detecting animal vocalizations in passive acoustic recordings while relying on far fewer labeled examples. It combines masked audio pretraining with self-training to improve robustness and transfer across recording sites. The approach aims to make large-scale biodiversity monitoring more practical where expert annotation is costly.
A new arXiv preprint addresses the limited ability of large audio-language models to pinpoint when specific sounds occur within a recording. The authors propose adding frame-level grounding during training so these models can localize audio events more precisely rather than only describing clips in broad terms. The work targets fine-grained temporal understanding, a known weak point for current audio-language systems.
A new arXiv paper introduces CRAF, a cross-view residual-aware fusion approach for detecting synthetic or converted speech. The method targets generalization to spoofing attacks not seen during training, a common weakness in deepfake audio detection. It builds on pretrained speech and audio models.
A new arXiv preprint examines how procedural audio should be scaled when used as a data source for learning transferable audio representations. The authors also ask whether training choices originally developed on natural audio still hold when the source is procedurally generated. The work aims to clarify design principles for procedural audio pre-training, an area the authors say lacks settled guidance.
Researchers present LoSATok, a tokenizer designed to serve both audio understanding and generation tasks within a single framework. The work argues that understanding benefits from high-level semantic features while generation needs both semantic and acoustic detail, and existing unified tokenizers encode both in high-dimensional spaces. LoSATok instead uses a low-dimensional representation to handle cross-domain audio.
Researchers propose BGM2Pose, a method that estimates a person's 3D pose without physical contact by using ordinary music playing nearby as an active sensing signal. The approach aims to avoid the intrusive chirp signals and impractical setups used by earlier acoustic pose-estimation systems. A revised version of the paper is available on arXiv.
A new arXiv preprint describes a multimodal dataset covering crossmodal correspondences between music and taste, a topic tied to intangible cultural heritage. The authors apply dataset normalization and perceptual validation so that computationally modeled links between sound and flavor hold up against human perception. The work targets applications in museums, exhibitions and gastronomic tourism.
Researchers present DuoTok, a tokenization method that separates vocal and accompaniment streams while keeping source information intact for multi-track music generation. The approach aims to balance acoustic detail, sequence modelability, and cross-track structure that existing codecs trade off against each other. A revised version of the preprint is now available.
A preprint describes an active noise cancellation approach designed for open-ear smart glasses, where the usual in-ear error microphone cannot be used. Conventional ANC depends on measuring residual sound at the ear canal, which the authors say is not feasible for this form factor. The work is posted on arXiv as a cross-list replacement and has not been peer reviewed.
Researchers introduce PitchFlower, a flow-based neural audio codec designed to separate pitch information from other audio content. The method applies a training perturbation in which F0 contours are flattened and randomly shifted, encouraging the model to disentangle pitch from the rest of the signal. This yields explicit, adjustable pitch control when synthesizing or compressing audio.
A new arXiv paper addresses voice-cloning fraud in which just a sentence or two of a genuine multi-speaker conversation is swapped for synthetic audio. Rather than giving one real-or-fake verdict for an entire recording, the proposed method pinpoints the exact time spans of fake speech without needing labelled examples of the targeted fakes.
A new paper studies how human labelers disagree about the precise onset and boundaries of laughter when annotating audio. The authors show that this disagreement follows systematic patterns rather than acting as random noise, challenging the common practice of scoring temporal laughter localization against a single reference annotation. They argue that evaluation protocols should instead account for the structured nature of annotator disagreement.
Deep spatially selective filters deliver high-quality, real-time speech enhancement for stationary speakers whose directions are known, but they struggle when sources move. A new arXiv paper introduces an autoregressive guidance scheme built on Bayesian speaker tracking that lets these filters follow moving speakers from only their initial positions. The approach preserves the efficiency and enhancement quality of the underlying architecture in dynamic scenes.
Researchers have introduced PRISM-Bench, a diagnostic benchmark for text-to-audio-video generation models that places the audio modality at the center of evaluation. The paper argues that prior benchmarks tend to treat sound as a minor add-on to video quality metrics or test it separately from the combined audiovisual output. The new benchmark is intended to give a fuller picture of how generative systems handle audio together with visuals.
A new arXiv paper examines ways to keep machine learning models from forgetting previously learned classes when trained incrementally on sound event classification. It evaluates architectural and regularization strategies using the FSD50K dataset and related benchmarks. The work is a research investigation rather than a released product.
A new arXiv paper addresses hallucination in large audio-language models, where models sometimes let language priors override what is actually heard in the audio. The authors propose a method that adaptively selects perturbations for contrastive decoding, a training-free approach, arguing that existing techniques rely on crude perturbations such as masking or added noise. The work aims to improve how reliably these models ground their outputs in acoustic evidence.
A developer known as Makestreme used neural-network-based compression to shrink an MP3 file so its data fits across several QR codes printed on a single page. The codes can be scanned and decoded to play the track without any internet connection.