Paper Archive

Sygyt is a Tuvan style of biphonic singing in which a low vocal drone is sustained while a high harmonic is selectively amplified in the 1–3 kHz region. Copy-synthesizing this effect remains challenging for articulatory models, since it requires fine control of narrowly focused resonances that standard low-dimensional tract parameterizations cannot easily reproduce. We address this problem with a differentiable Kelly–Lochbaum waveguide augmented with a sublingual second source, cubic B-spline tract parameterization, and spatially varying learnable damping, optimized end-to-end by gradient descent from audio. On 20 segments from two independent sygyt datasets (5 singers, 10 pitches), the proposed model reduces log-spectral distance by 30–38% relative to an articulatory baseline, with the largest gains concentrated in the overtone region. Cepstral-envelope analysis further shows more accurate recovery of the merged formant structure characteristic of sygyt production. The model also outperforms a DDSP harmonic-plus-noise baseline with direct per-harmonic spectral control, suggesting that explicit acoustic structure is a useful inductive bias for overtone-singing copy-synthesis.

Download

Artificial reverberation can be produced either by convolving a signal with a measured room impulse response (RIR) or by synthesizing it with a parametric algorithm such as a Feedback Delay Network (FDN). The former reproduces a captured space faithfully but is costly to run and offers no control over its acoustic properties, while the latter is efficient and editable but hard to match to a specific room. In this paper we bridge the two by fitting a fully differentiable FDN to a measured RIR through gradient descent. The proposed network uses sixteen delay lines at a sampling rate of 48 kHz and trains all of its components jointly, including the delay lengths, the feedback matrix, the early-reflection taps, and a set of attenuation filters that control the frequency-dependent decay.

Download

This work analyzes CLAP audio embeddings through a probing framework, studying the encoding of reverberation (RT60), loudness (LUFS), spectral content (SC), and relative pitch (RP). Results show that all attributes are reliably recoverable from CLAP embeddings, with RT60, LUFS, and RP approximately linearly encoded, while SC requires non-linear probes. The identified patterns generalize across eight additional audio foundation models.

Download

A clipping prevention method is proposed for first- and second-order all-pass filters with time-varying coefficients. Unlike conventional anti-clipping or declipping approaches, the method operates directly on the coefficient dynamics and does not rely on assumptions about internal energy evolution, by just asking that the input signal is not already clipping. The core idea is to control the deviation between the output of the time-varying filter and that of an equivalent static all-pass structure with constant coefficients. By adaptively limiting this deviation at runtime, the output is constrained below a prescribed clipping threshold (typically unit magnitude). The method is active only during short transients where clipping would occur, after which the coefficients are released to reach their target values. This preserves the integrity of the input signal and the numerical properties of the all-pass filter. Experimental results confirm the expected behavior even in scenarios where energy-preserving all-pass structures exceed the clipping threshold, suggesting the proposed approach as a practical solution for robust dynamic filter implementations with limited additional computational cost, suitable especially for embedded digital audio processing hardware.

Download

This work introduces a novel framework for Concatenation-Driven Convolution (CDC), unifying concatenative synthesis and real-time convolution into a single integrated audio processing paradigm. While concatenative synthesis has traditionally been used for corpus-based sound generation and convolution has served as a largely static filtering technique, the proposed approach reconceptualizes impulse responses (IRs) as dynamic, navigable sonic material. In the CDC framework, a corpus of audio segments is analyzed using perceptual features and organized via a self-organizing map (SOM), enabling intuitive, gesture-based traversal of a structured timbral space; the resulting concatenative output is treated as a continuously evolving impulse response and injected directly into a partitioned convolution engine. Its central technical contribution is single-engine frequency-domain kernel interpolation: rather than crossfading the outputs of two convolution engines, the FFT-domain kernels of the current and target IRs are interpolated within a single engine, preserving the internal convolution state across IR transitions and avoiding the warm-up energy loss inherent to dual-engine crossfading.

Download

Sound effects (SFX) datasets and libraries often employ distinct tagging schemes, taxonomies, and metadata structures. This creates challenges for research on SFX classification and generation because incompatible taxonomies lead to siloed datasets that might require individualized approaches, result in non-comparable outcomes, and prevent data merging strategies. We propose a modular dataset relabeling framework that adopts the Universal Category System (UCS), an industry-standard hierarchical taxonomy for sound effects, as a shared structural foundation. This open-source framework enables us (i) to convert tags of existing datasets to UCS with a rule-based multi-stage pipeline and conflict resolution to achieve high automatic conversion rates, (ii) to suggest a stratified dataset split for the new labels, and (iii) to combine multiple datasets. To showcase the practical utility, we introduce the EnvSound-UCS dataset, a publicly available unified UCS-compliant dataset of environmental sounds with 58,057 sound clips from three sources: AudioSet, FSD50K, and ESC-50.

Download

We present a simulation-trained, non-iterative estimator for Task A of the 1st DAFx Parameter Estimation Challenge. Each unnormalized plate-reverb impulse response is summarized by amplitude, spectral, and decay descriptors, and an ensemble of tree regressors estimates the six target parameters in one pass. Across two independent synthetic validation sets, the normalized models outperform the training-set mean and an earlier raw-regression baseline. On a shared set, the final ensemble also outperforms a single run of the official default PSO at substantially lower inference cost. Since the official labels are hidden, parameter accuracy is measured on simulator-matched data, and the released responses support only audio-side consistency checks. The estimator returns point estimates without uncertainty.

Download

We describe our submission to Task A of the 1st DAFx parameter estimation challenge. The task is to recover the six physical parameters of a simulated metal-plate reverberator—its dimensions and material properties—from a single impulse response (IR). We treat this as a black-box optimization: candidate parameter sets are fed to the simulator and scored by a loss against the target IR. The method has two stages. The first uses CMA-ES, an evolutionary optimizer, to recover five of the six parameters, comparing IRs under an amplitude-normalized loss. Amplitude normalization makes the search robust but discards the cue to the sixth parameter, the plate’s surface density; a second stage therefore estimates it alone, with a ternary search on the un-normalized loss. As the choice of loss strongly affects the search, we select it beforehand, and analyze why compression in the common multi-scale spectral loss degrades recovery. Finally, we test our method on a validation set of 50 IRs, discuss a pathological failure mode, and ablate to justify having two different stages instead of a unified CMA-ES search.

Download

A parametric modal reverberator is presented in which synthesis parameters are derived from a large, curated corpus of room impulse responses (IRs). The collected responses are subjected to modal decomposition, yielding per-mode frequencies, damping coefficients, and residue amplitudes, together with a short early-reflection finite impulse response (FIR) filter. From the decomposed data, a feature table is constructed per IR comprising standard acoustic indices, per-band damping and density statistics, amplitude distributions, and FIR descriptors—50 variables in total. Six acoustically meaningful user controls are selected; since these exhibit substantial pairwise correlations across the corpus, they are orthogonalised via principal component analysis (PCA) prior to regression.

Download

Spatial Room Impulse Responses (SRIRs) are fundamental to immersive audio rendering and have become a key focus of recent machine learning research in acoustics and auralization. Due to the high computational cost of direct convolution, spatial audio systems commonly employ artificial reverberation algorithms. However, these approaches often fail to accurately reproduce the spatial, temporal, and spectral characteristics of early reflections, leading to notable deviations from measured SRIRs. This paper presents a comprehensive framework for the analysis and efficient resynthesis of SRIRs captured with Spherical Microphone Arrays (SMAs). The proposed method accounts for hardware-induced artifacts, including scattering and spatial aliasing. Early reflections are reconstructed using a parametric approach based on the Herglotz analysis method, while late reverberation is synthesized using a Directional Feedback Delay Network (DFDN) with optimized filter-attenuation and correlation-matching. The proposed framework produces signals whose spatial correlation and Energy Decay Relief (EDR) closely match those of measured SRIRs, demonstrating its effectiveness for both real-time spatial audio rendering and realistic dataset generation for machine learning applications.

Download