AES Europe 2026

High-fidelity vocal processing is frequently compromised by
sibilance, a phenomenon characterized by stochastic
high-frequency energy that presents unique dynamic range
challenges. While traditional de-essing techniques often
rely on static frequency bands, they fail to account for
inter-speaker variability; changing dynamics. This
project presents an adaptive real-time de-essing
application, developed using the JUCE framework, which
automatically detects; suppresses sibilant frequencies.
The proposed methodology integrates a derivative-based
frequency tracking algorithm to estimate the spectral
centroid without the computational overhead of the Fast
Fourier Transform (FFT). This is coupled with a dual-path
envelope detection system; a relative threshold logic to
distinguish sibilance from the wideband signal.
Additionally, a dynamic harmonic exciter is implemented to
restore high-frequency presence during non-sibilant
periods. Objective spectral analysis confirms the system's
ability to selectively attenuate energy in the 6–11 kHz
range while maintaining spectral transparency;
minimizing artifacts.

This paper presents the production of a fully volumetric
audiovisual music composition with six degrees of
freedom, featuring a dance performance. The project
realizes the artistic potential enabled by recent
technological
advancements in volumetric video capture; spatial audio
rendering. The interdisciplinary production team
consisted of music composers, dancers, sound engineers,;
experts in 4D Gaussian Splats. An existing 3D body
scanner consisting of 112 cameras was used to capture a
dance performance in high definition video. For the
visual scene, custom 4D Gaussian Splat algorithms were
developed; employed to create the dynamic model.
Additional static 3D Gaussian Splats were captured with the
same scanner; integrated into the scene in Unity.
The acoustic scene is dynamically binauralized via SPAT
Revolution, depending on the position of the listener
in the virtual space. Audio; video scenes are run on
separate PCs, synchronized via OSC; presented via
commercially available head-mounted displays (HMDs).
Audiences report a high level of immersion at the initial
presentation at an exhibition event. A detailed evaluation
is planned in the near future. Furthermore, a unified
application for both visual; audio scenes is planned in
order to reach a wider audience.

Mix preparation, the foundational stage encompassing
technical, musical,; organisational tasks preceding
creative mixing, remains under-examined despite
professional acknowledgement. This study investigated
whether preparatory effectiveness operates through
compensatory relationships, where excellence in one
dimension offsets weakness in another, or through threshold
requirements demanding adequacy across all dimensions
simultaneously.

Nine professional audio practitioners each prepared three
sessions from a pool of nine multitrack recordings spanning
diverse genres. Nine engineers (with partial overlap) then
evaluated the resulting twenty-seven preparations across
five dimensions derived from Phase 1 practitioner
interviews: Session Organisation, Signal Integrity, Musical
Refinement, Processing Boundaries,; Workflow
Facilitation. Professional 'adequacy' was established at a
4.0 threshold based on practitioner consensus regarding
preparations they would 'work with' versus 'send back'.

Results revealed consistent non-compensatory patterns:
exceptional performance in isolated dimensions failed to
compensate for failures elsewhere. One practitioner
achieved perfect Workflow Facilitation (5.00) yet overall
inadequacy (3.43) due to Signal Integrity failure (2.50).
Another achieved strong Musical Refinement (4.75) whilst
Workflow Facilitation collapse (1.75) produced a
below-threshold outcome (3.49). These patterns held across
all inadequate sessions. No track produced exclusively
adequate or inadequate outcomes, confirming source material
did not determine success.

The findings challenge three assumptions: that
practitioners can specialise; compensate, that education
can sequence skills for later integration,; that
intelligent systems can optimise tasks independently.
Preparatory adequacy requires meeting threshold standards
across all dimensions concurrently, with implications for
professional hiring, curriculum design,; AI-assisted
tool development.

Historically, music has developed primarily as a frontal
phenomenon, thus limiting the expressive; perceptual
potential related to sound space. The recent development of
immersive audio systems opens new creative possibilities by
expanding the artistic action space from a narrow frontal
area to a complete sphere around the listener. The
Ambisonic system (Scene-Based Audio), together with
Object-Based formats; hybrid solutions, represents
fertile ground for creative experimentation; the
redefinition of workflows in the field of spatialized sound.
In this new context, what is the role of the sound
engineer, as an electroacoustic interpreter, in immersive
musical artistic creation?
The research is based on a multidisciplinary analysis that
combines an in-depth study of current immersive audio
technologies; their performance, with observations of
existing compositional; production approaches.
Additionally, a comparative study is conducted on the
design choices of the sound engineer as an interpreter,
investigating workflows, emerging musical semantics,
available tools,; the recovery of the historical
repertoire.
Particular attention is paid to the experiment aimed at
investigating a correlation between the position of a sound
; an emotional trigger in the listener.
New directions emerge in the creative role of the sound
engineer, who goes beyond the mere technical aspect to
become an integral part of the compositional;
interpretative process, harmonizing the relationship
between technique; art.

The low-frequency performance of cross-talk cancellation
(CTC) systems is fundamentally limited by the condition
number of the plant matrix, which indicates the robustness
of the inverse system in the absence of regularisation.
This condition number, in turn, depends on the relationship
between loudspeaker spacing, listener distance,;
acoustic wavelength.
This paper derives a simple approximate expression for the
low-frequency limit of CTC performance, defined for a given
maximum affordable condition number as a function of these
parameters. The increase in condition number is also shown
to be directly related to the increase in array effort
relative to the minimum achievable array effort. The
formulation is derived for a centered listener; can be
extended to the case of off-center listener positions,
demonstrating the method's applicability to
listener-position-adaptive cross-talk cancellation systems.

Mashup is a distinctive form of music composition which
integrates elements from existing songs to create a
cohesive audio experience. The digital music landscape,
with various audio processing tools; sharing platforms,
has facilitated the creation; propagation of mashups by
musicians, remixers, audio engineers,; automated
systems. While most prior research; studies focus on
mashups created by combining elements from individual audio
tracks, typically using pop songs, there exists other types
of mashups; for example, by incorporating phrases from base
melodies into a new arrangement. In this study, we examined
listener enjoyment ratings for this type of mashup,
utilizing well-known Western classical melodies. A
listening test was conducted to assess whether variations
in pitch, tempo,; familiarity with the source material
correlate with enhanced enjoyment. This paper presents our
preliminary findings, with plans for future studies;
additional survey responses to strengthen the results;
uncover insights for crafting more engaging classical
mashups.

This paper presents a multitrack dataset designed to
support music production research; education, including
machine learning techniques such as automatic mixing;
source separation. The dataset comprises a cohesive 10-song
indie album (indie rock/folk), with separate stems for
individual instruments, such that each song has between 13
; 35 individual tracks (stems). For each song, three
versions of each stem are provided: the raw unprocessed
stems, a dry mixed version (processed but without
reverberation or delay effects),; a full mixed version.
Additionally, each song includes two final master formats:
stereo; immersive 7.1.4. This album-format dataset
enables studies of mix consistency across a thematically
aligned collection of songs, as well as stereo upmixing to
immersive formats,; contains far more stems per song
than traditional four-stem datasets. To illustrate an
example usage of the dataset, the MEGAMI automatic mixing
model is used to produce a mix for two songs. The results
are analysed in comparison to the raw (unmixed); human
mixed versions. The dataset is made open-access; free to
download.

The MPEG-I Immersive Audio standard for Virtual;
Augmented Reality (VR/AR) audio with six degrees of freedom
(6DoF) was completed in November 2025 by the MPEG Audio
group (ISO/IEC JTC 1/SC 29/WG 6)). It offers compressed
representation of virtual audio scenes as well as an
efficient; acoustically sophisticated rendering to both
head-tracked binaural headphones; loudspeaker setups.
The latter is a unique feature among VR/AR audio
specifications; enables convincing reproduction of
conventional stereo, surround; 3D material with a large
sweet area in home entertainment setups for a single
tracked user, without the need for a head-mounted
display. This paper describes the technology of MPEG-I
Audio listener-tracked loudspeaker rendering as a
stand-alone application with a special focus on practical
considerations for optimal room calibration.

Dialogue intelligibility is a fundamental aspect of audio
post-production. Ensuring speech clarity in complex sound
mixes remains challenging across different playback
systems. Selective auditory attention plays a central role
in how listeners track dialogue in busy mixes, so small
changes in spectral or spatial structure can influence
perceived clarity in unexpected ways. This study
investigates the effectiveness of psychoacoustically
informed techniques, equalisation; spatialisation, in
reducing auditory masking; improving the clarity of
dialogue. The listening test was completed on participants’
own playback systems, which reflects typical domestic
viewing conditions; aligns the study with real-world
listening environments. The techniques were tested
individually; in combination to assess their impact.
Results show that equalisation was more effective than
spatialisation in reducing masking, while their combination
produced a significant improvement in intelligibility,
clarity,; reduced interference. The effectiveness of
these methods varied between the two groups of clips,
suggesting that their application should be adapted to the
specific acoustic context of each scene.

Sound plays a critical role in virtual reality (VR),
shaping attention, narrative comprehension, emotional
engagement,; experiential plausibility under conditions
of embodiment; user agency. Although a growing body of
research addresses VR audio techniques, perceptual effects,
; sound taxonomies, existing approaches remain fragmented
; largely descriptive. In particular, they do not provide
a unifying, VR-specific account of how sound meaning;
emotional intent are operationally linked to user agency
; non-linear narrative progression. This paper presents a
narrative review of selected literature spanning game audio
frameworks, immersive sound design, narrative theory,;
plausibility-related research in games; VR. Through
synthesis of these perspectives, the review identifies a
conceptual gap in current research, namely the absence of a
VR-specific, agency-coupled sound design framework for
structuring sound meaning; emotional intent in support
of experiential plausibility as users actively shape events
in interactive VR environments.

Most of music contents available are stereo which cause
inadequate spatial treatment; listeners feel
disconnected from the music, failing to transport them into
the intended sonic environment. Insufficient separation
between instruments can lead to an unbalanced mix, where
certain elements dominate others; disrupt the overall
harmony. Instruments may appear flat; confined to a
narrow area, reducing the sense of dimensionality in the
mix. Stereo audio offers limited spatial information,
restricting its adaptability to immersive sound
environments. This research presents a novel approach for
converting stereo audio into a personalized immersive
experience by leveraging object-based audio rendering,
sound stage of listener; surround speaker capability.
The proposed system separates audio signals into individual
objects (such as instruments or vocals); dynamically
maps these objects to specific speakers based on
personalized preferences; spatial configurations. This
method improves audio localization; enhances the
listener's engagement by delivering a tailored auditory
experience.

This paper presents a systematic literature review on
inverse synthesis; sound matching, which focus on
predicting synthesizer parameters to recreate a target
audio waveform. Automating this process using machine
learning is impeded by distinct technical challenges: many
to one mappings where different parameter settings produce
the exact same sound, the non-differentiability of
commercial black box synthesizers, a scarcity of musically
structured training data,; a lack of standardized
perceptual metrics. Existing approaches are categorized
into non-differentiable synthesizer methods, utilizing
evolutionary algorithms; deep learning, incorporating
techniques to bypass gradient limitations such as neural
proxies or generative models. In contrast, differentiable
synthesizer methods, enable the integration of audio loss
functions into training pipelines via custom signal
processing environments. The analysis identifies a critical
reliance on spectral representations for evaluating
perceptual similarity, given that parameter based metrics
frequently fail to align with human hearing. The findings
indicate that while deep learning has reduced inference
times, the field lacks a unified production solution.
Future progress requires the establishment of standardized
benchmarks to evaluate models, the implementation of novel
advancements in generative models not yet applied to this
problem,; the development of hybrid architectures to
simultaneously address these distinct technical challenges.

The bone-conducted occlusion effect (OE) is a major source
of acoustic discomfort for users of hearing aids, earbuds,
earplugs,; related devices. Conventional objective OE
measurements rely on in-ear microphones in human subjects,
which are time-consuming, invasive,; difficult to
control during product development. The aim of this paper
is to present a new artificial ear, specifically designed
for objective OE measurements under bone-conducted
excitation, coupled with a finite element analysis (FEA)
model developed in COMSOL Multiphysics. Both the model;
the artificial ear demonstrate good agreement regarding the
sound pressure found at the tympanic membrane for a
conventional dome at shallow, medium; deep insertions.
The validated FEA model is then used to perform a virtual
test of the bone-conducted objective OE for different
occluding devices, including plastic; foam earplugs;
a conventional closed dome for hearing aids. This is to
investigate the relative contributions; phases of the
ear-canal; device surfaces govern the resulting occluded
sound pressure. The proposed artificial ear; modeling
approach provide a controlled; repeatable platform for
studying the OE; for evaluating occluding devices during
the development process.

Digital Audio Signal Processing has long enabled precise
analysis of musical instrument behavior, supporting digital
sound synthesis. In parallel, physical modeling has evolved
into a mature synthesis; simulation technology capable
of running in real time, coupling vibro-acoustic models
with perceptual control interfaces. Over the last decade,
advances in machine learning have begun to transform both
ends of this pipeline. Instead of relying solely on
analytical DSP methods, we are increasingly able to learn
impulse; frequency responses, infer parameters,;
drive synthesis models directly from data. This broader
transition from classical DSP to *AI Audio Engineering*
brings not only new algorithms but also new workflows,
evaluation practices,; deployment contexts for musical
acoustics.

Two demonstrators illustrate this shift. *First*,
measurement-driven studies of musical instruments can
constrain model architectures; reduce parameter search
spaces. The measurement-derived priors can inform both
classical modeling; data-driven neural surrogates.
*Second*, real-time physical modeling integrated into XR
environments highlights how haptic control, perceptual
feedback,; spatial audio can create convincing virtual
instruments suitable for experimentation, pedagogy,;
performance.

These demonstrators motivate an AI Audio Engineering
workflow in which measurement, modeling, learning,;
perceptual evaluation form a continuous loop, to enable
immersive XR experiences, rapid prototyping of novel
instruments,; new modes of digital lutherie. The
approach invites collaboration across acoustics, DSP,
spatial audio,; AI Audio Engineering: an emerging
discipline that considers audio models as deployable,
maintainable,; continuously improvable artifacts
governed by data, inference, evaluation,; lifecycle
operations.

Accurate; efficient measurement of sound pressure levels
around the ears of occupants in cars is essential for
objective evaluation of basic sound quality; automotive
audio features such as personal sound zones; active
noise control. In this paper, the uncertainties of sound
pressure measurements obtained with 5 commonly used methods
are compared, which are the AES 6-microphone method, the
single-microphone method, the two-microphone method with
occupants presented, the head-and-torso simulator method,
; the human binaural method. Measurements were conducted
in the front-right seat of a 4-door electric Sedan, using
either all car body loudspeakers or a pair of headrest
loudspeakers driven by a two-channel uncorrelated pink
noise to generate an average sound pressure level of 70 dBA
in the seat. Each method underwent 3 complete
install–measure–remove cycles, a total of 54 recordings
were collected,; the standard deviation of the measured
average sound pressure levels was adopted to quantify
measurement uncertainty. The test results show that all the
5 methods have good repeatability; low uncertainty below
200 Hz; above 15 kHz, but have large uncertainty between
200 Hz; 15 kHz. The AES 6-microphone method demonstrates
the best repeatability with the lowest uncertainty across
most frequency resolutions,; its maximum uncertainty in
1/3 octave bands is less than 2.0 dB for sound pressure
measurements in the car. Therefore, the AES 6-microphone
method is recommended for use in engineering comparison;
reporting.

Artificial reverberation is a fundamental process in music
production; audio post-production. However, the large
; highly interdependent parameter spaces of modern
reverberation algorithms make the identification of
perceptually optimal configurations difficult, particularly
when attempting to minimize audible artifacts. This paper
presents a knowledge-driven framework for reverberation
parameter optimization that evaluates candidate
configurations using rule-based audio quality constraints
derived from perceptual; signal-processing principles.
The system automatically detects; prevents common
artifacts including spectral obfuscation, clipping, spatial
collapse,; ringing phenomena. Instead of relying on
data-driven training procedures, the proposed approach
employs declarative reasoning to model audio engineering
knowledge; systematically constrain parameter
exploration. Experimental evaluation demonstrates that the
framework successfully reduces artifact occurrence across
diverse audio material while maintaining computational
feasibility. The results suggest that knowledge-based
reasoning can provide an interpretable; controllable
alternative to data-driven optimization strategies in audio
signal processing.

Linear loudspeaker parameters are often estimated via
fitting of transferfunctions, under the assumption of
linearity. This paper investigates the corruption of the
measurement caused by nonlinearities in the system;
presents a new; improved method for separating the true
linear response from the nonlinear components by analyzing
a sequence of measurements done at different levels. The
method is improved by analyzing the influence of the chosen
measurement levels as well as the measurement time at each
level; presents numerically optimal values for the most
typical cases of nonlinear behaviour. While the influence
of noise; nonlinear distortion can be eliminated
completely in the case of finite orders of nonlinearities
on the system, the method is also shown to provide improved
accuracy in the more realistic case where all orders are
present but only a finite number of them dominate.

Automotive audio is challenging for a variety of reasons.
The acoustic environment is noisy, the geometry is complex
with many reflecting surfaces,; there are several
listening positions of interest. While digital signal
processing can to some degree alleviate some of the
associated issues, there often is a need for specialized
waveguides that directly affect the sound propagation from
the transducers. However, with the desired objectives being
quite intricate; involving on-axis pressure,
directivity, beam width,; possibly other metrics, the
design process is highly non-trivial. A strategy based on
acoustical topology optimization is presented here,;
where a tweeter waveguide to be mounted in the dashboard is
optimized towards certain objective functions.