Acoustics on Matias Di Bernardo

Acoustic modal response optimization for small rooms with genetic algorithms

Thu, 26 Jun 2025 00:00:00 +0000

This work was developed in the context of the course Instruments and Acoustic Measurements at UNTREF Argentina.

Introduction

Control rooms and critical listening environments often suffer from uneven low-frequency acoustic responses. These irregularities, caused by a clustered distribution of vibrational modes, produce “colorations” that hamper accurate sound evaluation. Traditionally, criteria such as those by Bonello, Bolt and Louden have been used to optimize the geometry of rectangular rooms and improve modal distribution. However, these methodologies do not consider the influence of complex boundaries nor the positions of source and listener.

This work presents an open-source tool developed in Python/FEniCS that addresses these limitations. The software uses geometric optimization by brute force over finite element models (FEM) to find room dimensions and contours that provide a more uniform modal distribution.

Theoretical Framework and Classical Criteria

Low-frequency behavior in an enclosure is dominated by standing waves, or normal modes, which are characterized by pressure nodes and antinodes. Axial, tangential and oblique modes — whose frequencies depend on the room dimensions — can produce coloration problems when they cluster.

Classical design criteria, such as those of Bolt, Bonello and Louden, focus on avoiding modal clustering and propose optimal geometric ratios for rectangular rooms. However, these approaches have a major limitation: they do not consider crucial factors such as the position of the sound source and the receiver, and they are restricted to simple geometries.

Methodology and Software Development

The developed tool combines a two-stage optimization process.

Initial Search: First, the software performs a rapid search on rectangular parallelepipeds using the classical modal superposition (MS) method to identify the most promising initial geometric proportions.
Refinement and Optimization: Then, it refines the search by generating random planar contours with enforced symmetry and applies the Frequency-Domain Finite Element Method (FD-FEM) to evaluate the acoustic merit of complex geometries. This method is more accurate than modal superposition for non-rectangular geometries.

To quantify acoustic performance, a combined figure of merit is used: the Mean Sound Field Deviation (MSFD). This metric integrates two key parameters:

Magnitude Deviation (MD): Measures how flat the frequency response is at a specific position.
Spatial Deviation (SD): Measures the variation of magnitude across the listening area.

The tool includes a graphical user interface (GUI) in PyQt5 that allows the user to define dimensions, margins, and source/receiver positions, and to visualize results and optimized geometries.

Results and Conclusions

Case studies on three reference control-room volumes showed MSFD improvements of up to 5 dB compared to the baseline design. The results demonstrate that:

Impact of Margins: As the available design space for optimization increases, better results are obtained, improving the overall response by up to 3 dB. This improvement is observed mainly in the Magnitude Deviation (MD).

Complex Geometries: Increasing the number of walls in a complex geometry produces solutions superior to simple rectangular parallelepipeds, with a mean difference of 1.3 dB in the merit factor. The optimization process does not yield a single solution but a variety of geometries that present a minimum MSFD.

Comparison with Traditional Criteria: A complex optimized geometry outperformed rooms dimensioned according to classic criteria by Bolt, Louden and Cox. Although these criteria are effective and computationally free, the software’s ability to model complex geometries and consider the locations of sources and receivers provides a superior condition.

The study concludes that the software is an effective tool for modal acoustic optimization. Future improvements are suggested, such as implementing more advanced optimization algorithms — for example, genetic algorithms — to reduce computation time and increase process efficiency.

A detailed analysis of the development of this algorithm is available in the following paper.

Acoustic measurement at Usina del Arte

Fri, 09 May 2025 00:00:00 +0000

Introduction

This measurement was part of the course Instruments and Acoustic Measurements of the Sound Engineering program at UNTREF. The acoustic parameters obtained from the impulse response are essential to evaluate the behavior of an enclosure. This report presents a comprehensive characterization of the main auditorium of the Usina del Arte, a cultural center in Buenos Aires. The building, originally a 20th-century power plant with a distinctive Florentine-industrial style, was transformed with an acoustic design that sought a natural and balanced quality without the need for amplification. A decoupled structure (box-in-box) was implemented for isolation and interior treatment with materials such as guatambú wood, diffusive surfaces, and a suspended acoustic reflector. The objective was a reverberation time of approximately 2 seconds and an even distribution of early lateral reflections for an enveloping sensation.

Measurement

The characterization was carried out on June 9, 2025, during which a total of 162 impulse responses (monaural and binaural) were recorded. Data were captured from 27 microphone positions and 3 source positions. An on-site survey of the auditorium was also performed to analyze its constructional characteristics and a perceptual analysis was conducted.

Prior to the measurements, a room model was created in EASE 4.3, which estimated a volume of 15,700 m³ and a Schroeder frequency of 22.1 Hz. Background noise was measured at eight positions to evaluate the isolation, confirming a signal-to-noise ratio greater than 40 dB. The microphone arrangement was based on the room’s symmetry to obtain a detailed mapping.

More images from the measurement process:

‹

›

Processing

Recordings were processed to obtain the impulse responses and various parameters were calculated following the ISO 3382-1 standard:

Reverberation time: $T_{20}$, $T_{30}$ and EDT.
Clarity: $C_{50}$ and $C_{80}$.
Strength (G): Difference in sound pressure level between the hall and an anechoic reference condition.
Lateral Fraction (LF): Proportion of sound energy perceived from the laterals.
Direct/reverberant ratio (D/R).
Intelligibility: The Speech Transmission Index (STI) and the Articulation Loss of Consonants (%Alcons) were calculated.
Stage support: $ST_{Early}$ and $ST_{Late}$, to assess acoustic conditions for musicians.

Various commercial software tools were used, such as the Aurora Acoustical Parameters plugin and the EASERA software, and additional parameters were computed with specific Python scripts.

Results

The results show that the auditorium behaves adequately for a concert hall, but with areas for improvement:

Reverberation time: The global average was 1.92 s. However, notable variations were observed at low frequencies, where the floating stage acts as a resonator.

Clarity and Intelligibility: Clarity values for speech are below recommended thresholds, and intelligibility issues were identified in certain zones.
Sound Strength (G): The sound strength level shows a low variation considering the auditorium’s dimensions.

Lateral Fraction (LF): Values exceed recommendations, suggesting that most of the sound energy comes from the laterals. This may be related to the large number of diffusers.
Background Noise: The room presents a noise level higher than recommended for a symphonic venue (NC-35 vs. NC-20), likely due to the ventilation system.

Sound Diffusion: Repetition of a single sequence of diffusers reduces their effectiveness, producing a lobed behavior instead of stochastic diffusion.

Key improvements are proposed, such as reducing background noise, optimizing sound diffusion with non-periodic sequences, and balancing the spectral response by correcting low-frequency absorption.

Conclusions

We were able to effectively characterize the auditorium and apply most of the theoretical topics covered in class to a practical experience. The full report of this work with all results and measurement details can be found in the following report.

The effect of denosing on TTS

Fri, 22 Nov 2024 00:00:00 +0000

This study was conducted in the context of the class Laboratorio de Acústica at UNTREF. I chose this topic because it aligns with research I have been pursuing as part of the group Intercambios Transorgánicos. The class assignment involved conducting a subjective study using a survey to explore the relationship between objective and subjective variables.

In my research group, I have been investigating how denoising algorithms affect Text-to-Speech (TTS) systems trained on low-quality recordings. The focus is on Rioplatense Spanish, a regional accent with limited high-quality data. Within this context, it was natural to combine both tasks and perform a subjective test on the impact of denoising algorithms on TTS systems.

Overview

The key points of this investigation are:

Evaluation of three denoising algorithms: Wave U-Net, HiFi-GAN, and DeepFilterNet.
Use of both subjective (CMOS) and objective metrics (PESQ, STOI, MCD).
Insights into resource-efficient TTS model development for underrepresented accents.

Methodology

Algorithms: Wave U-Net, HiFi-GAN, and DeepFilterNet evaluated with the FastPitch TTS model.
Dataset: Subset of the ArchiVoz collection (15 minutes of noisy audio).
Testing: CMOS subjective test and objective metrics (PESQ, STOI, MCD).
Participants: 24 valid responses, including both experts and non-experts.

Key Findings

DeepFilterNet Performance:
- Achieved the highest CMOS score, reflecting the best subjective quality.
- Demonstrated significant improvements in TTS output despite mixed correlations with objective metrics.
Objective Metrics Analysis:
- PESQ and MCD showed limited correlation with subjective preferences.
- STOI scores were consistent across algorithms, indicating preserved intelligibility.
Algorithm Comparisons:
- DeepFilterNet: Superior subjective evaluations, moderate MCD.
- Demucs: Comparable to DeepFilterNet in PESQ but lower subjective scores.
- Wave U-Net: Poor subjective and objective performance.
Subject Expertise:
- No significant differences were observed between expert and non-expert evaluations in subjective testing.

Implications

Efficiency: Advanced denoising methods like DeepFilterNet can enhance TTS systems without requiring high-quality recordings.
Limitations: Objective metrics like PESQ and MCD are insufficient standalone indicators of subjective TTS quality.
Future Work:
- Expand datasets and noise levels for more robust analysis.
- Explore TTS systems trained jointly with denoising algorithms.

Conclusions

This work concludes that preprocessing with DeepFilterNet significantly improves TTS performance, with a 1.1 CMOS score increase. These findings underscore the importance of algorithm selection in optimizing low-resource TTS systems. Additionally, I gained valuable insights into subjective evaluations and the statistical analysis required to draw meaningful conclusions from data.

All the information for this study can be found in the academic report.

Building and design of a personal loudspeaker

Wed, 20 Nov 2024 00:00:00 +0000

“BassAdo”: A Semi-Portable Low-Cost Home Speaker

This project is part of the Electroacoustics II course at UNTREF within the Sound Engineering program. The task was to design a speaker system from scratch by applying the theory and concepts explained in class. The project was developed over the entire semester, with various stages to complete and present in reports. The speaker is intended for use in large spaces, potentially outdoors, to play music in a social gathering setting. It was named BassAdo to blend the Argentine tradition of “asado” (a typical barbecue gathering) with the word “bass,” emphasizing the speaker’s low-frequency performance.

Design

The goal was to design an accessible home audio system that allowed exploration of topics discussed in the course. The design aimed to emphasize bass response, characteristic of commercial systems, prioritizing low-frequency bandwidth extension over minimal group delay and system time control.

Regarding the transducers, the team had access to Yharo-brand units, which are classified as non-professional, consumer-grade, and suitable for automotive or home systems. The impedance response of the units was evaluated, and an 8” woofer was selected for low frequencies, along with two 4” units for mid/high frequencies.

Measuring the Thiele-Small parameters of the speakers revealed a high Vas (Equivalent Suspension Acoustic Volume), necessitating a large cabinet volume for proper control. To address this, and given the availability of two 8” woofers, the team opted for an isobaric speaker configuration, acoustically coupling the woofers to improve control and reduce cabinet size. Additionally, the cabinet was designed as vented to enhance low-frequency response.

The Thiele-Small parameters were obtained using the software REW. With these parameters, simulations were performed in Basta! to optimize the design for the desired response. A key focus was tuning the port’s resonance frequency to achieve strong low-frequency performance. The transducer had an fs of 45 Hz, and the port was tuned to 40 Hz by adjusting the tube length and cabinet air volume.

Based on these results, a 3D model of the cabinet was created using SolidWorks, and the design was used to cut the materials for construction.

Details of this process are documented in the following design report.

Construction

The wood was cut according to the 3D model, and the cabinet was assembled.

‹

›

As shown in the images, rock wool was added as an acoustic absorber. Measurements revealed this was excessive (the port resonance was overly damped), so some rock wool was removed to achieve the desired result.

Measurement and Calibration

Frequency response and directivity measurements were conducted in the university’s laboratory using the following equipment:

Powersoft M50Q amplifier
Earthworks M50 microphone
RME Fireface UCX audio interface
OUTLINE ET250-3D turntable

Using this setup and the Arta software, the acoustic response of individual transducers was characterized (useful for crossover filter simulation). Vertical and horizontal directivity responses were also evaluated to determine the best orientation for use. Frequency response graphs for both transducers were generated.

All measurements and in-depth analysis are included in the following measurement report.

Crossover Filter Design

Finally, the crossover filter stage was designed. Using the previous measurements, data were uploaded to VituixCad to calculate the simulations. The goal of the crossover filter was to achieve a pleasant frequency response for music playback and to enhance low frequencies. Vertical polar response uniformity was also a priority.

Since the power stage required an active supply, an active crossover filter with a Sallen-Key topology was implemented. The number of filters was defined based on space and cost, and adjustments were made in the software to achieve the desired response. For example, the low-frequency driver used the following configuration:

Where:

F1: High-pass fs=30 Hz | Q=0.67
F2: Low-pass fs=480 Hz | Q=0.5
F3: Notch filter at 220 Hz
F4: Notch filter at 400 Hz

Before building the filter, the proposed configuration was tested with a digital filter to practically evaluate the system’s response. Details of this section are provided in the following crossover filter report.

Conclusions

This project allowed us to apply theoretical concepts in practice and gain a deeper understanding of the development and challenges involved in designing an electroacoustic system.

Theather Acoustic Design

Wed, 22 Jun 2022 00:00:00 +0000

This project is the final assignment for the class Acoustics and Psychoacoustics II, where we were tasked with redesigning an existing auditorium. The goal was to apply the theory covered in class to create an acoustically optimized auditorium. For our project, we chose to redesign the Royal Albert Hall in London. This was particularly challenging due to the auditorium’s vast dimensions, which make it difficult to ensure that sound reaches all spectators equally.

Redesign Main Ideas

The redesign aimed to preserve the original concept of the auditorium, including its large volume and extensive seating capacity, while introducing critical changes to improve its acoustics. Although the primary focus was on acoustic enhancement, the redesign also considered other essential factors, such as sightlines and appropriate seat distribution.

Despite the intent to maintain the auditorium’s original dimensions, its volume proved too large to achieve an optimal reverberation time. To address this, the redesign introduced an intermediate ceiling to reduce the spherical ceiling’s volume, and the main seating area was reduced. These changes helped create a better reverberation time in the room, as illustrated in the cross-section below.

Building Details and Regulations

To ensure a feasible and functional redesign, the following key aspects were carefully considered:

Seat distribution
Corridor spacing
Sightline optimization
Stage comfort

Acoustic Treatment

Acoustic treatment was the most critical part of this study and focused on two main aspects: reflections and reverberation time.

Reflections

Analyzing reflections is essential for the audience’s acoustic experience. The original Royal Albert Hall features a spherical ceiling that centralizes reflections, creating undesirable acoustic effects. To mitigate this, the redesign incorporated an intermediate ceiling with a specific geometry designed to distribute reflections evenly across the audience.

The staggered ceiling design ensures adequate reflections for all seating rows. In the main balcony, two reflections were specifically addressed to compensate for the lower sound pressure level (SPL) caused by the large distance from the stage, as shown in the image below.

Lateral reflections were also optimized through adjustments to the stage geometry and the walls of the lateral balconies.

Additionally, the redesign sought to minimize the Initial Time Delay Gap (ITDG) across different audience locations.

Materials and Reverberation Time

The redesign adhered to recommendations from Acoustic Absorbers and Diffusers to achieve a balance between absorption, diffusion, and specular reflections. Reflective materials were used for the ceiling and parts of the lateral balconies to ensure effective specular reflections. To lower the reverberation time (RT), materials with higher absorption coefficients were applied to other surfaces.

Using the selected materials and the Sabine equation, we calculated the auditorium’s estimated RT. The resulting reverberation time for different frequencies is shown below:

The calculated mid-frequency RT is 2.51 seconds. While this is slightly above the recommended maximum of 2.4 seconds for optimal acoustics, it is acceptable given the auditorium’s large volume.

3D Modelling

We rendered the redesigned auditorium using SketchUp software. Below are some of the visualizations:

‹

›

Conclusions

Redesigning the Royal Albert Hall to improve its acoustics while retaining its original essence presented significant challenges. The project required innovative solutions to address acoustic issues without compromising the hall’s iconic design. Although some changes were necessary, the final result demonstrates a thoughtful redesign that enhances acoustics while preserving the auditorium’s historical character. This project also deepened our understanding of acoustics and auditorium design principles.

A detailed description of this project can be found in the following article.