Speech Separation

Real-time music source separation is validated on desktop CPUs and GPUs. Does any published system fit the embedded audio hardware it targets? On a commercial audio DSP (2 MB SRAM, 2.07 GMAC/s measured), none does, and the constraints eliminate different models: memory rules out the 16-51 M parameter TasNet/X-UMX family, per-frame compute rules out RT-STT, needing 5.5x the available MAC rate. Parameter count predicts neither: weight reuse spans 1x to 345x. We then build one that fits. Training on continuous rather than block-padded convolution context proves essential: a model scoring 3.93 dB block-wise otherwise collapses to silence within 2 s frame-by-frame. A gated complex FIR deep filter adds a latency knob, gaining 0.38 dB even when strictly causal. It reaches 4.70 dB cSDR on MUSDB18-HQ and runs in 10.43 ms of an 11.6 ms hop, 0.5-0.7 dB behind systems that do not fit.

Distant speaker diarization remains challenging due to difficult acoustic environments, varying numbers of speakers and overlapping speech. Model-driven methods are proposed to exploit the speech source features in multi-channel recordings that help diarization. This paper generalizes a neural model that jointly learns to perform blind source separation and diarization over speech mixtures (neural FCASA) with heavy-tailed models. The popular Gaussian distribution has been applied for variance modeling in the original source separation model, which we replace with two families of heavy-tailed models (Leptokurtic Generalized Gaussian distribution and Student's t distribution) to better capture the heavy-tailedness in speech signals. Thanks to the Gaussian scale mixture model, we are able to unify the proposed method and the original one under the same form of learning objective. Our experiments show consistent large improvements in Diarization Error Rate (DER) and Jaccard Error Rate (JER) compared to the baseline on various corpora.

Conv-TasNet has served as a strong baseline for time-domain speech separation, and many studies have extended it with advanced architectures such as dual-path networks, U-Nets, and attention mechanisms. However, these methods often introduce high computational cost and complexity, limiting their deployment in resource-constrained scenarios. To address this issue, we propose eConv-TasNet, an efficient variant of Conv-TasNet that improves both effectiveness and efficiency without relying on resource-intensive modules. The proposed model consists of a group-wise early-splitting (GES) module and a multi-group feature aggregation (MGFA) module. GES generates discriminative speaker embeddings at intermediate stages, while MGFA progressively aggregates these group-level representations for refined mask estimation. Experimental results show that eConv-TasNet reduces model size by 22.4%, accelerates inference by 18.9%, and improves SI-SNRi by 14.0%-28.0% across three public benchmarks. Moreover, it achieves competitive performance compared with state-of-the-art methods while requiring significantly fewer parameters and lower inference cost. These results demonstrate a favorable efficiency-effectiveness trade-off for edge deployment.

This paper proposes a general framework for stable and effective iterative audio separation with mixture consistency by extending source separation models to a multi-input multi-output (MIMO) configuration. In the field of audio separation, mixture consistency is an essential property for many applications that require accurate phase and timbral information of target sources. While iterative approaches such as diffusion models achieve perceptually superior results in speech enhancement or user-guided target source separation tasks, most existing methods focus on single-step separation with a single-input single-output (SISO) or single-input multi-output (SIMO) configuration through architectural improvements, since mixture-consistent audio separation is generally regarded as a regression problem that admits a unique solution. By extending these architectures to a MIMO configuration, we introduce iterative prediction without compromising the architectural advantages or the characteristics of mixture consistency. We conduct a comprehensive ablation study of combining the framework with discriminators and extending it to a generative model. Experimental results demonstrate significant performance improvements when applying the proposed framework to state-of-the-art separation models.

The separation of cardiovascular sounds is a critical task in biomedical signal processing. In this paper, we introduce XVAE-WMT1, an unsupervised explainable generative AI algorithm combining a variational autoencoder (VAE) with explainable AI (XAI), wavelet-based inputs, a post-hoc output mask, and temporal consistency (TC) loss. Unlike existing supervised and VAE-based methods that rely on Short-Time Fourier Transform (STFT) and ignore latent interpretability, XVAE-WMT requires no paired clean recordings and integrates a Continuous Wavelet Transform (CWT) front-end for superior time-frequency localization. We assessed the latent space interpretability via different metrics, with SHAP (SHapley Additive exPlanations) enabling dimensionality reduction to the top 75% of latent features while preserving separation quality. Evaluated across two datasets using Signal-to-Distortion Ratio (SDR), Signal-to-Interference Ratio (SIR), and Signal-to-Artifacts Ratio (SAR), XVAE-WMT attains 26.8 dB SDR, 32.8 dB SIR, and 28.6 dB SAR.

Generative modeling provides a flexible way to model mixture-conditioned source distributions, but iterative diffusion and flow matching models are costly for long music signals. This paper studies joint vocal-accompaniment separation through latent flow matching, where a pretrained variational autoencoder (VAE) maps mixtures and sources into a compact latent space and a flow matching model generates vocal and accompaniment latents jointly. The proposed framework decouples semantic separation from acoustic velocity prediction through a Separation Encoder and a Velocity Decoder. To reduce sampling cost, we further apply latent adversarial post-training inspired by Flow2GAN for few-step generation. Experiments show that latent adversarial refinement can improve perceptual and separation metrics under a reduced sampling budget.

Long speech separation typically employs a segment-separation-stitch paradigm where recordings are divided into short segments, processed independently, and stitched together. Its challenge lies in predicting cross-segment permutations. This paper proposes a training-free dynamic clustering approach for cross-segment permutation alignment using speaker embedding reference pools. The method predicts the permutation using the cosine similarity between current segment embeddings and the reference pools. The approach updates reference pools by retaining the most representative speaker embeddings based on their overall cosine similarity with existing references. As a plug-and-play post-processing module compatible with existing separation models, the proposed method demonstrates superior performance compared to existing methods on dense and sparse long speech scenarios, particularly in challenging sparse scenarios with extended utterance gaps, and further shows robustness to speaker count estimation errors in unknown speaker count scenarios.

Audio-visual speech enhancement under real-world conditions remains challenging due to unreliable visual inputs and the lack of large-scale training data with realistic acoustic conditions. Existing approaches usually fuse visual features directly into the separation network, making them vulnerable to degraded visual signals. In this paper, we present DAVE, a decoupled audio-visual enhancement framework for real-world speech separation. Firstly, to address the data scarcity issue, we construct DAVE-Corpus, a large-scale training corpus with 219,411 mixtures generated from public meeting corpora through combinatorial acoustic augmentation. Then, we introduce a progressive multi-objective optimization strategy to jointly improve speech separation, intelligibility, speaker identity preservation, and perceptual quality. We further develop a certified selective enhancement chain that applies scene routing, GAN-based denoising, and loudness normalization only within the no-reference partition, guaranteeing non-degradation of reference-based metrics. Experimental results on the Real-World Audio-Visual Speech Enhancement Challenge demonstrate the robustness of DAVE under both real-world mixed scenarios and visual degradation conditions.

Humans have the remarkable ability to engage in spontaneous informal conversations and selectively attend to individual speakers while filtering out competing speech from nearby conversations. This "cocktail party" scenario still presents severe challenges to speech recognition systems. The CHiME-9 MCoRec task provides a testbed where systems must recognize groups of speakers and transcribe each of their conversations from audio-visual input. In this work, we analyze a diverse set of systems, representing different design directions for addressing the cocktail-party scenario, where the best system achieves up to 57% relative error reduction. We identify three main strategies: (1) explicit or implicit audio-visual target speech separation, (2) improved audio-visual speech recognition for each target speaker, and (3) the use of large language models to group speakers into conversations and enhance conversational consistency. Our analysis shows that these directions address complementary failure modes of the cocktail-party problem, and that high speech overlap alone does not explain performance differences, challenging the common assumption that overlap is the primary source of difficulty in cocktail-party recognition.

We benchmark eleven audio classification methods: five task-aware closed-set LLMs (four Gemini models plus open-weight Kimi-Audio-7B-Instruct), four fixed-vocabulary taggers (YAMNet, PANNs, Whisper-AT, and SSLAM), a zero-shot audio-text model (CLAP), and an audio-grounded LLM (BAT). We evaluate them on a closed-set sound-source identification task over 2,242 clips spanning 23 fine-grained classes and 11 categories. Since these methods differ fundamentally in how they receive the task and how outputs are scored, we group them into four evaluation tiers rather than one leaderboard, reporting macro Precision, Recall, F1, and false-negative rate per tier. The best model, Gemini-3.1-Pro-Preview, reaches 85.6 percent category-level F1 and 56.7 percent fine-grained F1. Kimi-Audio is competitive for its size, reaching 67.5 percent category-level F1 and 32.9 percent fine-grained F1, but fails to answer 1.6 percent of samples. SSLAM and CLAP match or exceed the best closed-set model at the category level without seeing the candidate list, but fall behind at the fine-grained level. Analyzing the Gemini models' chain-of-thought across 8,968 responses, we find that response length does not predict accuracy, an apparent "holistic judgment beats detailed analysis" effect is better explained as a difficulty confound, and wrong answers are stated confidently 92 to 100 percent of the time. We report full per-class confusion matrices and metrics for all eleven methods, identify the structural error modes behind most of the accuracy loss between granularities, and give practical guidance for choosing among these method families.

Automatic Drum Transcription (ADT) is commonly formulated as a direct mapping from a music mixture to symbolic drum events. While effective for transcription, this formulation discards the acoustic stems that are useful for editing, remixing, and production. We revisit an alternative separate-and-detect formulation, where a drum source separation front end first produces five editable drum stems, and a fixed onset detector then converts each stem into symbolic events. The separator is built on a five-stem latent diffusion model that jointly generates kick, snare, toms, hi-hats, and cymbals in a compact VAE latent space. We further study two training-only auxiliary branches--an onset branch (OB) and a timbre branch (TB)--which shape the separator during learning but are discarded at inference. Trained on synthetic drum multitracks and evaluated on MDB Drums and ENST-Drums, the proposed pipeline consistently improves over a strong U-Net-based drum separation baseline in overall transcription F1. It also outperforms a representative end-to-end ADT system on kick and snare F1 under our evaluation protocol, while additionally providing separated audio stems. The ablation results show that OB gives the most stable transcription gains, whereas TB changes the trade-off between reconstruction, acoustic stem quality, and onset detection. These results suggest that generative drum demixing can serve not only as a source separation model, but also as a practical front end for interpretable drum transcription.

Most Music Source Separation (MSS) models do not generalize well to live music recordings because they are trained on studio recordings alone, disregarding the venue acoustics, the speaker system's response and audience noise. We propose to bridge this gap by providing and training a model on two novel datasets. First, we present CrowdioSet: a noise dataset comprising 4800 real ambience tracks from Freesound and synthetic sing-alongs for the vocals in MUSDB18 and MOISESDB datasets, generated from zero-shot singing voice conversions. CrowdioSet enables effective audio denoising for live recordings, resulting in superior separation both in objective and subjective evaluations. Second, we introduce PaRIRset, a stereo impulse response dataset captured across 40 professional concert venues using a microphone array. Our results show that adding PaRIRset RIRs increases the performance of a MSS model compared to using real RIRs from Speech Enhancement tasks alone. We make the examples, code, model weights, PaRIRset, and CrowdioSet freely available to the public.

Despite the rising popularity of immersive audio, binaural music remains underexplored in music information retrieval (MIR), particularly regarding the task of music source separation (MSS). While existing stereo MSS models can process binaural audio, they often degrade the spatial quality of the separated stems and undermine listener immersion. Through a perceptual study comparing binaural and stereo MSS outputs, we evaluate how well objective spatial distortion metrics correlate with human perception. Our findings reveal varied agreement between these metrics and human judgment, highlighting a lack of reliability when used to evaluate binaural music tasks. Specifically, we find that Interaural Time Difference (ITD) estimation is highly sensitive to noise and separation artifacts. In evaluating two alternative ITD estimation methods, we uncover a critical trade-off between robustness and accuracy, particularly for narrow-band instruments like bass. These results underscore the need for accurate, interpretable spatial metrics designed for binaural music to develop models that preserve source localization and listener immersion.

Music Source Separation (MSS), the task of recovering individual sound components (stems) from a polyphonic mixture, is central to applications ranging from karaoke and remixing to audio restoration and content production. The separation quality depends on engineering decisions across the entire pipeline: model choice, training data preparation and augmentation, loss function and metrics choice, training configuration, validation, and post-processing. This paper presents MSST (Music-Source-Separation-Training) - a universal open-source framework for MSS tasks, which unifies training, validation, and inference for a broad range of modern demixing model families under a single, configuration-driven interface. The framework supports various model architectures, data preprocessing and augmentations, multiple loss functions and evaluation metrics, which helps with fast iterations and ablation studies. Additionally, the framework supports a range of practical techniques that improve separation quality, such as sliding-window inference with cross-fading, test-time augmentation, model ensembling, and fine-tuning via Low-Rank Adaptation (LORA). Our ablation studies demonstrate improvements of MSS using the above techniques. By consolidating these components into a reproducible, YAML-configurable framework, MSST lowers the barrier to systematic experimentation and enables rapid iteration from idea to verifiable result.

Transformers with global attention capture long-range dependencies but can miss the fine-grained local continuity crucial for speech separation. We propose TF-MossFormer, a time-frequency transformer that combines local and global attention to jointly model short- and long-range contexts for monaural speech separation. At its core is a content-aware sliding-window attention mechanism that dynamically adapts receptive fields for stronger local interactions, avoiding the rigidity of static convolutions. Unlike time-domain chunk-based methods, TF-MossFormer leverages the 2D spectrogram to model structure along both time and frequency axes. Convolutional gating between attention layers further improves feature selection and information flow. TF-MossFormer achieves SI-SDRi of 22.6, 24.0, and 24.4 dB on WSJ0-2Mix with 5.9M, 16.9M, and 25.4M parameters, respectively, outperforming prior approaches.

State-of-the-art speech enhancement models benefit from large-scale labeled datasets, whereas singing voice separation models suffer from limited available training data. To address this limitation, we formulate singing voice separation as domain adaptation from speech enhancement to singing voice separation. We investigate two fine-tuning strategies: full fine-tuning and parameter-efficient fine-tuning using Low-Rank Adaptation (LoRA) on a discriminative and a generative model. Models with either adaptation strategy outperform the same architectures trained from scratch by 0.29-1.8 dB in Signal-to-Distortion-Ratio. Full fine-tuning yields the highest singing voice separation performance, but catastrophic forgetting degrades speech enhancement performance. LoRA fine-tuning achieves competitive singing voice separation performance while preserving the original speech enhancement capability with only 6-12% additional parameters compared to the base speech enhancement model. Furthermore, the generative model shows improved generalization to an unseen test set. The results demonstrate that adapting pretrained speech enhancement models is an effective strategy for training singing voice separation models in data-scarce scenarios.

Single-channel speech separation remains challenging for real-world deployment due to source permutation ambiguity, sampling variability of generative models, and the difficulty of processing long recordings with chunk-wise inference. We address these issues with a conditional flow-matching-based method that produces an ordered two-source output conditioned on the mixture. A frozen speaker encoder defines the source order during training and is reused at inference for biometric best-of-$N$ candidate selection and chunk-level channel alignment. We evaluate separation quality on Libri2Mix benchmark using SI-SDR, PESQ, and ESTOI, and measure downstream impact using cpWER for automatic speech recognition and EER for speaker verification. The results show that the proposed Transformer U-Net variant is competitive with strong baselines in objective separation metrics and achieves the lowest downstream automatic speech recognition and speaker verification error rates in all evaluated settings.

Recent research expands beyond binary anti-spoofing with the emergence of Source Tracing, the task of identifying the specific generative origins of synthetic speech. However, current research often equates a "source" with its generative architecture. We propose redefining a source as a compositional tuple of Architecture, Training Data, and other training factors affecting the generated speech. We propose a framework using Structured Orthonormal Prototypes to minimize class overlap and intra-class variance. Our Subspace Partitioning strategy splits the embedding into architecture and data subspaces, while a residual subspace captures stochastic variability, enabling "compositional generalization" for novel factor combinations. This approach improves performance for partially seen sources and maintains robustness in fully open-set scenarios. MLAAD evaluations for Few-Shot open-set Identification show our approach significantly outperforms angular-margin baselines.

Designing effective feature extractors is critical for blind signal analysis tasks such as automatic modulation recognition (AMR), signal scheme recognition (SSR), and \color{black} signal structure parsing (SSP). In this work, we propose dual-channel neural network (DualNN) that efficiently exploits complex-valued signals through parameter sharing across IQ channels. Unlike traditional real-valued or complex-valued models, DualNN is a groundbreaking framework which shares the network parameters for processing the real and imaginary parts of the complex-valued signals, and is theoretically shown to reduce generalization error while preserving expressive capacity. Specifically, we propose a novel Transformer-based architecture to implement DualNN, called Dualformer. The Dualformer segments input signals into patch-level tokens and captures multi-granularity features, enabling robust performance across diverse signal analysis tasks. Furthermore, we conduct extensive experiments comparing Dualformer with three Transformer-based baselines and four conventional DL-based approaches. Results demonstrate consistent performance improvements on AMR, SSR, and SSP tasks. Besides, the modular design of DualNN allows it to generalize well to blind signal processing tasks such as blind source separation and low-SNR spectrum sensing. This work paves the way for a broader application of DualNN architectures in unsupervised and weakly supervised complex-valued signal analysis scenarios.

Recent advances in speech separation (SS) have led to compact front-end models with small parameter sizes, yet their high computational cost remains a major barrier for deployment on edge devices. To address this, we propose TF-MoE, a sparse Mixture-of-Experts (MoE) framework that enhances model capacity with almost no increase in inference cost. Our method introduces dynamic expert specialization in time and frequency dimensions through alternating time-wise and frequency-wise MoE modules, each dynamically selecting experts per frame or mel band. Built upon a mel-band-splitting Conformer backbone, TF-MoE achieves strong performance on SS tasks under low-compute settings. Experimental results demonstrate that TF-MoE consistently improves separation performance under computation cost constraints, outperforming BSRNN by +3.8 dB SDR on Libri2Mix with comparable 4.1 GMACs/s inference cost. This positions TF-MoE as a promising candidate for edge-device deployment.

In long-form multi-party conversations, highly imbalanced speaker activity and frequent overlap make it difficult to identify "who spoke when and what". Sliding-window continuous speech separation (CSS) mitigates sparse supervision, but often suffers from cross-window speaker inconsistency and residual crosstalk, which in practice requires diarization for reliable speaker attribution. Motivated by the stability of speakers' directions of arrival (DOAs) in meetings, we propose PATSE, a multi-channel Position-Aware Target Speaker Extraction front-end that uses DOA as a spatial prior to directly extract the speech of each target speaker. PATSE combines a DOA-guided spatial encoder and conditioner to generate speaker-attributed streams, from which speaker activity can be inferred via simple post-processing (e.g., VAD) without explicit diarization. Experiments on both replayed and real conversations show consistent ASR gains outperforming CSS and diarization-based pipelines.

We derive the linear union-of-subspaces (UoS) model for subspace clustering (SC) from the nonlinear mixture model (NMM) used in blind source separation (BSS) to represent a D-dimensional observation vector as an unknown multivariate nonlinear mapping of C latent variables. Assuming the mapping is differentiable up to an unknown order K, we approximate NMM by a K-th order Taylor expansion, yielding a model equivalent to the linear UoS framework underlying SC. This establishes that: (i) the smoothness order K corresponds to the unknown subspace dimension d; (ii) KC equals the number of anchors; and (iii) the sparsity of the representation vector equals K (i.e., d). These relationships enable estimation of bounds on subspace dimension, and that is validated on six benchmark datasets using five established SC algorithms. Established theoretical results are important for post-processing of self-representation matrices estimated by SC algorithms.

Music source restoration (MSR) requires jointly addressing source unmixing and the inversion of non-linear production effects. Current methods struggle to achieve accurate target signal reconstruction while maintaining semantic consistency. To address this limitation, we propose DTT-BSR+, a two-stage cascade MSR system that decouples distribution fitting from signal reconstruction into separate stages. A generative DTT-BSR separator in the first stage produces stems matching the prior of clean sources, and a modified Demucs network in the second stage enhances the first stage output using time-domain and multi-resolution spectral losses. DTT-BSR+ improves multi-mel signal-to-noise ratio (MMSNR) over the single-stage DTT-BSR across all stems, and surpasses the state-of-the-art X-LANCE MSR system on five stems. We also reveal through Fréchet Audio Distance (FAD) decomposition an implicit trade-off between signal reconstruction accuracy and semantic distribution fitting across stems.

We propose AugCodec, a low-bitrate disentangled neural speech codec that leverages data augmentation to decompose speech into three distinct components: semantic, speaker, and prosody tokens. Specifically, we employ tailored augmenta tion strategies to transform speech into distinct variants, each serving as input for extracting tokens that preserve the target attribute while suppressing others. This disentanglement strategy enables substantial reduction in token rate. Further more, we introduce an augmentation loss that aligns semantic encoder outputs between source and voice-converted speech, encouraging speaker-agnostic embeddings while mitigating the acoustic mismatch induced by voice conversion. Experiments on LibriSpeech test-clean demonstrate that AugCodec significantly outperforms state-of-the-art methods in both reconstruction quality and disentanglement, while operating at only 12.5Hz with three token streams.

Reference-based adaptive interference cancellation is evaluated for stereo audio recordings corrupted by real train noise and environmental background. The observed signal is modeled as a clean stereo program contaminated by an additive disturbance generated by an external acoustic source through unknown propagation paths. A second stereo recording, representing another filtered observation of the same physical noise source, is used as the reference input of a multi-reference recursive least-squares (RLS) estimator. The estimated train-interference component is subtracted from the noisy audio and followed by a finite-impulse-response low-pass postfilter. Three 74.01 s real audio sequences sampled at 11.025 kHz are processed under identical algorithmic parameters. Since clean ground truth is not available, performance is assessed with no-reference indicators: waveform behavior, Welch spectral estimates, RMS change, and residual normalized correlation with the reference. With 30 taps per reference channel, 15 anti-causal taps, and forgetting factor 0.999, the maximum reference correlation is reduced from 0.386--0.832 before processing to 0.011--0.016 after processing. The corresponding correlation-ratio reduction is approximately 30.6--34.1 dB, while the output RMS decreases by 1.8--4.8 dB depending on section and stereo channel. The results demonstrate that real train interference, including environmental acoustic effects, can be substantially attenuated when a correlated reference recording is available.

This paper proposes a novel confidence score guided incremental and speaker adaptive pseudo-labeling approach for semi-supervised elderly speech recognition. It facilitates higher-quality pseudo-label selection and progressive refinement, while also mitigating speaker heterogeneity. A confidence estimation module is designed to rank the reliability of untranscribed data, enabling a curriculum learning trajectory that progressively folds in unlabeled data subsets from high to low confidence. Speaker-specific characteristics are captured through speaker adaptive training with learnable prompts. Experiments on the English DementiaBank Pitt and Cantonese JCCOCC MoCA elderly speech datasets suggest that the proposed method outperforms the semi-supervised baseline using no confidence scores guided incremental or speaker adaptive pseudo-labeling by statistically significant word error rate (WER) or character error rate (CER) reductions of 1.45% and 2.27% absolute (6.21% and 6.98% relative).

Simultaneous speech-to-speech translation aims to enable near-real-time communication by minimizing latency, offering a compelling, real-time alternative to the high latency of consecutive translation. However, the excessive pursuit of low latency often results in fragmented chunk-wise speech. Consequently, listeners are subjected to an unnatural acoustic flow punctuated by frequent pauses, which could increase their cognitive load. To bridge this gap, we introduce a fluency-aware optimization framework designed to discover the sweet spot between the low-latency benefits of simultaneous translation and the natural flow of consecutive translation. Our framework minimizes inter-chunk silences by leveraging model-internal signals, including linguistic diversity and induced temporal variability in speech durations. Experiments on short- and long-form benchmarks show that our framework produces natural speech flow while maintaining competitive latency and translation quality.

Training neural networks (NNs) for speech enhancement (SE) in distant speech-capturing scenarios requires paired distorted and clean reference speech signals. While such data are often generated through simulation, the mismatch between simulated and real recordings significantly limits SE accuracy. To address this issue, we propose Close-to-Distant microphone Projection (C2D projection), a method that generates paired data from real recordings captured by close and distant microphones. C2D projection estimates an optimal projection matrix that transforms close-microphone inputs into clean reference signals aligned with distant-microphone recordings, while simultaneously performing denoising. We show this projection can be effectively realized using a variant of the Parametric Multichannel Wiener Filter (PMWF). Experimental results demonstrate that an NN trained with C2D-projected data outperforms the state-of-the-art Guided Source Separation (GSS) on the challenging CHiME6 dinner party ASR task under oracle diarization, when using the enhanced output from GSS as an auxiliary input to the NN.

Flow-matching transformers achieve strong audio separation, yet their attention dynamics are opaque. We adapt established causal-intervention principles into a deterministic, inference-time probing protocol for SAM Audio. Orthogonal probing uncovers a dual-pathway text-conditioning mechanism: additive injections control semantic identity, while cross-attention refines acoustic structure. We observe an asynchronous layerwise convergence: stable layers build temporal scaffolds early, whereas fast layers continue resolving artifacts during sampling. The model also attenuates temporal segmentation cues to maintain continuous-flow stability. Using these insights, we propose Layer-Selective Attention Caching (LSAC), a training-free acceleration method that caches attention in stable layers. Across acoustic complexities, LSAC cuts self-attention computation by about ~25% with negligible quality loss and yields up to 6.7x higher quality retention than naive step reduction.

While discriminative models for multi-channel speech separation excel in reference-based metrics, they often exhibit suboptimal human listening quality. To address this, we propose a novel MeanFlow-based one-step generative corrector (MeCo). MeCo learns a conditional average velocity field to map discriminative estimates directly onto the clean speech manifold in a single step. To maximize one-step generation performance, we introduce Data-Space Optimization (DSO). DSO integrates an $\mathbf{x}_r$-loss, which penalizes prediction errors on longer displacement intervals to serve as a generative objective for human listening quality, with an Endpoint SI-SDR loss that directly optimizes terminal signal fidelity. Experiments demonstrate that MeCo achieves state-of-the-art (SOTA) performance with minimal computational overhead, simultaneously achieving superior signal fidelity and human listening quality in both in-domain and out-of-domain scenarios.

The goal of single-channel source separation is to reconstruct $K$ sources given their mixture. In supervised settings where vast amounts of clean source data are available, this challenging, ill-posed problem has been addressed successfully by generative diffusion and flow-based prior models. However, access to such clean source samples is often limited, and even when available, supervised models are vulnerable to domain shifts. To bridge this gap, we present Separation via Unsupervised Remixing Flow (SURF), an unsupervised flow matching approach for source separation that learns directly from observed mixtures. This method relies on a novel combination of state-of-the-art supervised flow matching and regression-based self-supervised techniques. At a high level, starting from a teacher model, we utilize a "remixing" step to bootstrap the learning of a student flow model from the teacher's estimates. We provide insights into the objectives optimized by this approach and draw a novel connection to the Wake-Sleep algorithm. Empirical evaluations on image and audio benchmarks demonstrate that SURF establishes a new state-of-the-art, significantly outperforming existing unsupervised methods. See our demo page for examples. https://google.github.io/df-conformer/surf/

Hearing aids impose strict latency and power constraints that current DNN-based speech enhancement systems struggle to meet on embedded hardware. We characterize this gap by deploying both speech separation and denoising using the lightweight SuDoRM-RF++ architecture on the AMD-Xilinx Kria KV260, evaluated at FP32 and 16-bit fixed-point precision for each task. Across these configurations, first-sample latency tracks with on-chip parameter caching rather than arithmetic throughput, identifying data movement as the primary bottleneck. Precision reduction halves the model memory footprint without compromising objective speech quality. The fixed-point denoising accelerator achieves a first-sample latency of 9.7ms, meeting the 10ms clinical threshold, while speech separation reaches 16.0~ms. These measurements establish concrete resource requirements for embedded DNN-based speech enhancement and quantify the remaining gap to hearing aid deployment.

In conversational speech separation and recognition tasks, close-talk microphones are typically attached to each speaker during training data collection to capture near-field, close-talk mixture signals, in addition to using far-field microphones to record far-field mixture signals. Each such close-talk mixture exhibits a reasonably high energy level for the wearer and could intuitively serve as weak supervision for training far-field speech separation models directly on real-recorded far-field signals. However, they are not sufficiently clean for this purpose, as they often contain strong cross-talk speech from other speakers in addition to background noise. To address this, we propose cross-talk reduction (CTR), a task aiming to isolate the wearer's speech from each close-talk mixture, and a novel method called CTRnet, which can be trained directly on real-recorded pairs of close-talk and far-field mixtures to accomplish CTR. Building on CTRnet, we further propose pseudo-label based far-field speech separation (PuLSS), which uses CTRnet's estimated clean speech as pseudo-labels to train models for separating far-field mixtures. A key advantage of the proposed framework is that both CTRnet and PuLSS can be trained on real-recorded data from the target domain, addressing the generalization gap commonly observed when models are trained exclusively on simulated data. On the CHiME-6 dataset, our framework achieves state-of-the-art ASR performance under both oracle and estimated speaker diarization, surpassing all CHiME-{7,8} challenge submissions. To our knowledge, it is the first neural speech separation method that substantially outperforms guided source separation on real conversational "speech-in-the-wild" data.

Timbre transfer aims to modify the timbral identity of a musical recording while preserving the original melody and rhythm. While single-instrument timbre transfer has made substantial progress, existing approaches to multi-instrument settings rely on separate-then-transfer pipelines that propagate source separation artifacts and produce incoherent synthesized timbres across stems. This paper proposes MixtureTT, to the best of our knowledge the first system for flexible per-stem timbre transfer directly from a polyphonic mixture. Given a mixture and a separate timbre reference for each target voice, MixtureTT jointly transfers all stems to the specified instruments through a shared diffusion process. Modeling the dependencies across the per-stem content and cross-stem harmonic, the proposed joint stem diffusion transformer eliminates cascaded separation error, reduces inference cost by a factor equal to the number of stems, and yields more coherent multi-stem outputs. Despite operating under a strictly harder input condition, evaluations on the SATB choral dataset show that MixtureTT outperforms single-instrument baselines on both objective and subjective metrics demonstrating the necessity of dedicated multi-instrument timbre transfer over the naive separate-then-transfer pipelines. As a result, this work confirms that the cross-stem modeling is essential for mixture-level timbre transfer as the proposed joint setting consistently exceeds an equivalent single-stem ablation.

We propose a plug-and-play framework for speech enhancement and separation that augments predictive methods with a generative speech prior. Our approach, termed Stochastic Interpolant Prior for Speech (SIPS), builds on stochastic interpolants and leverages their flexibility to bridge predictive and generative modeling. Specifically, we decompose the interpolation dynamics into a task-specific drift and a stochastic denoising component, allowing a predictive estimate to be integrated directly into the generative sampling process. This results in a mathematically grounded framework for combining strong pretrained predictors with the expressive power of generative models. To this end, we train a score model using only clean speech, yielding a degradation-agnostic prior that can be reused across tasks. During inference, the predictor provides a deterministic drift that steers the sampling process toward a task-consistent estimate, while the score model preserves perceptual naturalness. Unlike prior hybrid approaches, which typically rely on architecture-specific conditioning and are tied to particular predictors or degradation settings, SIPS provides a unified framework that generalizes across predictors and additive degradation tasks. We demonstrate its effectiveness for both speech enhancement and speech separation using recent predictors such as SEMamba and FlexIO. The proposed method consistently improves perceptual quality, achieving gains up +1.0 NISQA for speech separation.

Evaluation of musical source separation (MSS) has traditionally relied on Blind Source Separation Evaluation (BSS-Eval) metrics. However, recent work suggests that BSS-Eval metrics exhibit low correlation between metrics and perceptual audio quality ratings from a listening test, which is considered the gold standard evaluation method. As an alternative approach in singing voice separation, embedding-based intrusive metrics that leverage latent representations from large self-supervised audio models such as Music undERstanding with large-scale self-supervised Training (MERT) embeddings have been introduced. In this work, we analyze the correlation of perceptual audio quality ratings with two intrusive embedding-based metrics: a mean squared error (MSE) and an intrusive variant of the Fréchet Audio Distance (FAD) calculated on MERT embeddings. Experiments on two independent datasets show that these metrics correlate more strongly with perceptual audio quality ratings than traditional BSS-Eval metrics across all analyzed stem and model types.

This paper proposes StrEBM, a structured latent energy-based model for source-wise structured representation learning. The framework is motivated by a broader goal of promoting identifiable and decoupled latent organization by assigning different latent dimensions their own learnable structural biases, rather than constraining the entire latent representation with a single shared energy. In this sense, blind source separation is adopted here as a concrete and verifiable testbed, through which the evolution of latent dimensions toward distinct underlying components can be directly examined. In the proposed framework, latent trajectories are optimized directly together with an observation-generation map and source-wise structural parameters. Each latent dimension is associated with its own energy-based formulation, allowing different latent components to gradually evolve toward distinct source-like roles during training. In the present study, this source-wise energy design is instantiated using Gaussian-process-inspired energies with learnable length-scales, but the framework itself is not restricted to Gaussian processes and is intended as a more general structured latent EBM formulation. Experiments on synthetic multichannel signals under linear and nonlinear mixing settings show that the proposed model can recover source components effectively, providing an initial empirical validation of the framework. At the same time, the study reveals important optimization characteristics, including slow late-stage convergence and reduced stability under nonlinear observation mappings. These findings not only clarify the practical behavior of the current GP-based instantiation, but also establish a basis for future investigation of richer source-wise energy families and more robust nonlinear optimization strategies.

Full-duplex dialogue audio, in which each speaker is recorded on a separate track, is an important resource for spoken dialogue research, but is difficult to collect at scale. Most in-the-wild two-speaker dialogue is available only as degraded monaural mixtures, making it unsuitable for systems requiring clean speaker-wise signals. We propose DialogueSidon, a model for joint restoration and separation of degraded monaural two-speaker dialogue audio. DialogueSidon combines a variational autoencoder (VAE) operates on the speech self-supervised learning (SSL) model feature, which compresses SSL model features into a compact latent space, with a diffusion-based latent predictor that recovers speaker-wise latent representations from the degraded mixture. Experiments on English, multilingual, and in-the-wild dialogue datasets show that DialogueSidon substantially improves intelligibility and separation quality over a baseline, while also achieving much faster inference.

Modern audio is created by mixing stems from different sources, raising the question: can we independently watermark each stem and recover all watermarks after separation? We study a separation-first, multi-stream watermarking framework --embedding distinct information into stems using unique keys but a shared structure, mixing, separating, and decoding from each output. A naive pipeline (robust watermarking + off-the-shelf separation) yields poor bit recovery, showing robustness to generic distortions does not ensure robustness to separation artifacts. To enable this, we study separation-aware watermarking in a controlled verification pipeline, where the separator is part of the detector and can be selected or optimized together with the watermarking system. Experiments on speech+music and vocal+accompaniment mixtures show substantial gains in post-separation recovery while maintaining perceptual quality.

Recent advances in automatic speech recognition (ASR) and speech enhancement have strengthened the common belief that cleaner audio should lead to more accurate transcription. In this work, we examine whether this assumption holds for modern zero-shot ASR systems. We conduct a structured empirical study of SAM-Audio as a preprocessing step for zero-shot transcription with OpenAI Whisper. Five Whisper variants are evaluated on noisy Bengali and English speech datasets. On the English dataset, SAM-Audio increases the average PSNR from 32.28 dB to 35.99 dB and achieves higher PSNR for 71.84% of the utterances. However, WER and CER increase in every evaluated model-dataset configuration. On the Bengali dataset, Whisper large-v3 WER increases from 65.83% to 77.35%, while CER increases from 24.13% to 34.74%. On the English dataset, Whisper base WER increases from 10.53% to 21.66%, while CER increases from 4.48% to 12.50%. Utterance-level analysis further shows that the degradation affects a substantial portion of the evaluated samples, although its severity varies across Whisper variants. These findings demonstrate that improved signal-level quality does not necessarily lead to better zero-shot ASR performance and that denoising can reduce recognition accuracy.

Text-guided sound separation enables flexible audio editing, assistive listening, and open-domain source extraction, but systems such as AudioSep remain too expensive for low-latency edge or codec-mediated deployment. Existing neural audio codec separators are efficient, yet largely restricted to fixed stems or closed taxonomies. We introduce CodecSep, a prompt-driven universal sound separation framework that extracts sources directly in neural audio codec latent space. CodecSep combines a frozen DAC backbone with a lightweight FiLM-conditioned Transformer masker driven by CLAP text embeddings, enabling open-vocabulary separation while preserving codec-native efficiency. Across dnr-v2 and five open-domain benchmarks, CodecSep consistently improves over AudioSep in SI-SDR, remains competitive in ViSQOL, and achieves clear gains in human MOS-LQS. Controlled analyses show that fine-grained prompts outperform coarse labels, and that explicit latent masking is substantially more effective than decoder-style latent generation in codec space. Qualitative diagnostics show that neural audio codec latents retain source-dependent structure, which CodecSep exploits mainly through channel-wise source-conditioned modulation. CodecSep also provides a practical code-stream deployment path. When audio is transmitted as neural audio codec codes, CodecSep maps codes to embeddings, separates directly in codec space, and outputs waveforms or re-quantized codes, avoiding the decode-separate-re-encode loop. In this regime, CodecSep requires only 1.35 GMACs end-to-end: about 54 times less compute than AudioSep in the same pipeline and 25 times lower separator-only compute, with much lower latency and memory. More broadly, CodecSep offers a blueprint for codec-native downstream audio processing.

Environmental sound recordings often contain intelligible speech, raising privacy concerns that limit analysis, sharing and reuse of data. In this paper, we introduce a method that renders speech unintelligible while preserving both the integrity of the acoustic scene, and the overall audio quality. Our approach involves reversing waveform segments to distort speech content. This process is enhanced through a voice activity detection and speech separation pipeline, which allows for more precise targeting of speech. In order to demonstrate the effectivness of the proposed approach, we consider a three-part evaluation protocol that assesses: 1) speech intelligibility using Word Error Rate (WER), 2) sound sources detectability using Sound source Classification Accuracy-Drop (SCAD) from a widely used pre-trained model, and 3) audio quality using the Fréchet Audio Distance (FAD), computed with our reference dataset that contains unaltered speech. Experiments on this simulated evaluation dataset, which consists of linear mixtures of speech and environmental sound scenes, show that our method achieves satisfactory speech intelligibility reduction (97.9% WER), minimal degradation of the sound sources detectability (2.7% SCAD), and high perceptual quality (FAD of 1.40). An ablation study further highlights the contribution of each component of the pipeline. We also show that incorporating random splicing to our speech content privacy enforcement method can enhance the algorithm's robustness to attempt to recover the clean speech, at a slight cost of audio quality.