Two years of SQAT: an open-source repository of sound quality metrics 

by Gil Felix Greco, Roberto Merino-Martínez, Alejandro Osses, Michael Lotinga, and Sabine C. Langer  

1. Introduction 

In modern sound engineering, sound quality (SQ) assessment is critical to evaluate and optimize the acoustical characteristics of products for which sound perception is a relevant feature. For example, the SQ of a vacuum cleaner is a parameter that can influence consumer purchasing decisions and, therefore, deserves as much consideration as its weight or power. Sound quality can be considered a perceptual attribute per se, for example, the quality of Sound A is better than of Sound B. Alternatively, SQ can reflect different qualities of sound, described by different perceptual dimensions, such as perceived magnitude or pitch, among others. 

In this context, literature suggests that SQ metrics established by psychoacoustic models are able to better represent loudness and short-term annoyance of sounds than conventional metrics based on the sound pressure level (SPL). Additionally, psychoacoustic models that describe other perceptual dimensions, such as tonality or roughness, offer alternative descriptors for sound analysis. Nevertheless, the calculation of psychoacoustic models is far from straightforward. While some models are standardized (e.g. ISO 532-1 model for loudness), implementing other models commonly adopted in the literature may be challenging, due to the lack of clear instructions. As a consequence, different implementations of the same model can lead to varying outcomes, highlighting how the lack of verified, well-documented, open-source implementations poses a challenge to reproducible research. 

This was the shared opinion between Roberto, Gil, and Alejandro when they met during the International Congress on Acoustics in October 2022 in Gyeongju (South Korea). All of us had some experience implementing some of these psychoacoustic models, but we lacked a proper tool containing readily-accessible and validated implementations that could be distributed among institutions for reproducible research. Therefore, we decided to employ a preliminary version of the Sound Quality Analysis Toolbox (SQAT) developed by Gil as a starting point and released it in May 2023 for the general public. A comprehensive description of the implementation and verification of the toolbox can be found in [1] and an illustrative case study application can be found in [2]. The most recent version includes implementations of the Sottek’s psychoacoustic hearing model standardised SQ metrics (ECMA-418-2), provided by Michael Lotinga, who joined the team of maintainers in 2025. The verification and implementation of the ECMA-418-2 metrics can be found in [3]. The live GitHub repository can be found in: https://github.com/ggrecow/SQAT; and v1.3 of SQAT in: https://zenodo.org/records/15148618

Figure 1 – Evaluation paradigms for sound quality analysis, highlighting SQAT as a tool providing objective measures for sound quality evaluation. The GitHub repository of the toolbox is available in the QR code or in  https://github.com/ggrecow/SQAT. 

In a nutshell, the models implemented in SQAT can be used as objective metrics to evaluate SQ. As illustrated in Figure 1, SQ analysis is typically conducted based on subjective and objective approaches. Ideally, SQ is quantified in listening experiments, where human perception is directly included on the evaluation process. From listening experiments, SQ metrics can be established by correlation of subjective evaluations with objective parameters associated with the sound stimuli (e.g., SPL, frequency or amplitude modulations). Thereby, objective metrics predicting the subjective evaluations are established, providing a fast yet representative way to quantify SQ without relying on time-consuming and expensive listening experiments. Nevertheless, such objective metrics are derived based on specific contextual and experimental variables, and as such, their general applicability to other cases must be ideally verified or performed carefully.    

In SQAT, a series of MATLAB codes implementing different sound metrics and psychoacoustic models that can be used as objective SQ metrics is provided. For each implementation, rigorous verification is conducted and transparently made available so that users can be aware of their limitations. This article provides a high-level overview of SQAT’s structure, functionalities, and its relevance for the acoustics community. 

2. Toolbox structure 

The SQAT repository follows the folder structure explained in Figure 2: 

Figure 2 – Folder structure and metrics available within SQAT v1.3. Adapted figure from [1]. 

psychoacoustic_metrics 

This folder hosts a number of algorithms implementing specific psychoacoustic metrics, which are by default computed using calibrated input sound signals, in Pascal units. The algorithms are named using the following convention: MetricName_FirstAuthorYear. At the moment of writing this article, SQAT v1.3 includes the following implementations: 

  • Effective Perceived Noise Level: metric commonly used in aircraft noise certification to describe perceived “noisiness” under the guidelines of the Federal Aviation Regulations, 14 CFR, Parts 36 and 91. 
  • Loudness: quantifies perceived sound magnitude. Models from ISO 532-1 (Zwicker) and ECMA-418-2 (Sottek) standards are implemented. 
  • Sharpness: measure of the high-frequency content of a sound. The more high-frequency content, the “sharper” the sound is perceived. This metric is implemented according to the DIN 45692 standard (Widmann weighting, or alternatively, Aures or von Bismarck weightings). 
  • Roughness: hearing sensation caused by modulation frequencies between 15 Hz and 300 Hz, which lead to the impression of a “rough” or “harsh” sound. The models of Daniel & Weber and ECMA-418-2 (Sottek) are implemented. 
  • Fluctuation strength: hearing sensation caused by slow fluctuations in loudness, with modulation frequencies up to 20 Hz, which give the impression of a “pulsating” or “fluctuating” sound. SQAT model by Osses et al. is implemented. 
  • Tonality: relates to unmasked single-frequency or narrowband components that emerge audible from the total sound. Both the Aures and ECMA-418-2 (Sottek) models are available. 
  • Psychoacoustic annoyance: Three models are currently available in SQAT for calculating psychoacoustic annoyance as a combination of the psychoacoustic metrics. 

Sound_level_meter

Contains scripts to obtain the time-dependent SPL using different frequency weightings (A, B, C, D, or Z) and time weightings (fast, slow, or impulse), and other metrics derived from it. By default, the input is assumed to be a calibrated sound signal, in Pascal units.

Examples

In order to facilitate the initial use of the algorithms, an exemplary script is provided for each metric as a template. As a general guideline, the metrics implemented follow the following structure:  

OUT = MetricName_FirstAuthorYear(insig,fs,additional_parameters) 

where: 

  • OUT is a structure variable containing the outputs relevant to the metric and some statistical values. 
  • insig is the input signal assumed calibrated and in Pascal units. 
  • fs is the sampling frequency of the input signal in Hz. 
  • additional_parameters may include factors like the choice between different methods to be used, the choice between free or diffuse field, etc. 

sound_files

Contains WAV files required to run any codes within the examples folder. 

utilities

Scripts that are complementary to any of the toolbox functions are provided here. The main utility scripts concern the calibration of the input signals and default input values for the implemented metrics.

validation

Contains scripts used to verify a particular algorithm. 

publications

Contains scripts to reproduce figures and/or tables of publications from the toolbox authors.  

3. Future work and community involvement 

Future features for SQAT in the long-term future may include: 

  • Despite being quite intuitive, non-programmer users could benefit from the addition of a graphical user interface (GUI) to SQAT. 
  • Since MATLAB is technically not freely available, translating SQAT to (or making it compatible with) other open programming languages, such as Python or similar, could increase its usability. For users of both platforms, the MATLAB Engine API for Python provides a bridge, enabling MATLAB-language functionality to be called from Python code. 
  • Expansion to include new sound metrics, such as impulsiveness or the speech intelligibility index. The inclusion of new metrics in the future is foreseen as long as enough verification and documentation related to their implementations are (or can be made) available. With this strong requirement, we will only publish implementations that are mature enough for general use, without necessarily requiring further updates after a code is released. 

We kindly encourage contributions, bug reports, and feature requests through the GitHub repository. The development team welcomes collaboration to expand SQAT’s capabilities and maintain its relevance.  

At the moment of writing this article, the use of SQAT has been documented in a total of 24 scientific publications. These are listed in the GitHub repository, since, as developers, we are very happy to know that SQAT has been useful. In case we miss any contribution, we kindly ask you to reach out to us so we can feature your work there. 

The team of maintainers are committed to carefully maintaining the toolbox in the long term, thus supporting its use as a reliable and readily accessible tool. Ensuring backward compatibility of the codes across SQAT releases is also a key requirement. However, small unforeseen issues will be fixed as quickly as possible. If that is the case, any changes will be clearly stated it in the documentation. For this reason, we recommend that users always inform which version of SQAT has been used, and to use the most up-to-date release of the toolbox. 

References 

[1] Felix Greco, G., Merino-Martinez, R., Osses, A., and Langer, S.C., “SQAT: a MATLAB-based toolbox for quantitative sound quality analysis,” 52nd International Congress and Exposition on Noise Control Engineering, 20-23 August 2023, Chiba, Greater Tokyo, Japan. DOI: 10.3397/IN_2023_1075. 

[2] Osses, A., Felix Greco, G., and Merino-Martinez, R., “Considerations for the perceptual evaluation of steady-state and time-varying sounds using psychoacoustic metrics,” 10th Convention of the European Acoustics Association (Forum Acusticum), 11-15 September 2023, Torino, Italy. 

[3] Lotinga, M.J.B., Torjussen, M., and Felix Greco, G., “Verified implementations of the Sottek psychoacoustic hearing model standardised sound quality metrics (ECMA-418-2 loudness, tonality and roughness),” 11th Convention of the European Acoustics Association (Forum Acusticum), 23-26 June 2025, Málaga, Spain.