Page 16 - Winter 2020
P. 16

of acoustic simulation methods into virtual reality systems (Vorlander, 2013, 2020). These types of systems can have real-time performance due to the advances in technology and have become paramount in the entertainment industry.
Additionally, virtual and augmented realities have been employed in training and as a diagnostic tool. In the past, there used to be latency or slowing down of simulations due to the huge amounts of data being generated. However, this is not as significant problem anymore given advances in computer technology. As a result, sound synthesis and production of indoor/outdoor surroundings can be com- bined with three-dimensional stereoscopic display systems through data fusion (e.g., Vorlander, 2020). The research and design applications have led to improved reality for video games and similar systems. The user experience is enhanced by adding accurately synthesized sound and allowing the listener to be able to move unrestrictedly, e.g., turn the head, to be able to perceive a more natural situation.
Moreover, the improved synthesis algorithms (e.g., Gao et al., 2020) can be used to provide more realistic conditions for psychoacoustic tests. Sound synthesis algorithms based on deterministic-stochastic signal decomposition have been applied to synthesize pitch and time scale modifica- tions of the stochastic or random component of internal combustion engine noise (Jagla et al., 2012). The method uses a pitch-synchronous overlap-and-add algorithm, used in speech synthesis, that exploits the use of recorded engine noise data and the fact that the method does not require specific knowledge of the engine frequency. The data-based method used for speech synthesis, noise analy- sis, and synthesis of engine noise just mentioned is similar to what is used in ML. Applications of ML seem to have no limits in the data-driven world of today.
ML methods are based on statistics and are excellent at detect- ing patterns in large datasets. Applications in acoustics are fertile ground for research into ML for things such as voice recognition, source identification, and bioacoustics (e.g., Bianco et al, 2019). With technologies like Alexa or Google Home, voice recognition investigations are needed to allow the technology to work with people having different accents or pronunciations or speaking different languages. The algo- rithms must utilize huge datasets of recorded voices to teach the computer system to “learn” based on input. Models are developed of voices pronouncing certain common words used for searching. Variations are compared statistically to the
model where the model can be improved based on additional inputs of data. The computer algorithm from the system using it essentially “learns” and incorporates that knowledge into its dataset. Although much of the research into ML and techniques are done in areas of computer science, the appli- cations of the methods into acoustics have driven some of the more recent advances. A major method of ML, called deep learning, based on artificial neural networks that work through several layers, train systems to do everything from synthesizing music to being able to perform better than the human ear for recognition (Hawley et al., 2020).
Summary and Conclusions
The large variety of methods and applications outlined here is hardly an exhaustive depiction of computational acoustics. Due to limitations in my knowledge and the space and time to do so, only a brief introduction to the field could be given. However, hopefully, I was able to make the case for the need for the field of computational acoustics and the variety of areas of application. The uses of computational methods have driven discovery and improved understanding in a variety of areas of acoustics including sound synthesis, voice recognition, modeling of acoustic propagation, and source identification. Several techniques have been used to aid in the design of new automotive technologies by modeling the mechanical interactions of structures with different moving parts and the fluids involved.
Several of these methods are not only being used in engi- neering acoustics, but they are also being employed for space design for concert halls and classrooms. This type of modeling has improved noise suppression in a variety of mechanical systems. Computational techniques are being used in modeling and simulation in signal process- ing to utilize ML methods in the investigation of acoustic source identification and classification. The methods are being applied to areas of animal bioacoustics to aid in species identification for population monitoring, avoid- ing direct interaction with the animals. The methods and applications of computational acoustics are only going to grow over years to come and have become a fruitful and rewarding area of research.
The opinions and assertions contained herein are my private opinions and are not to be construed as official or reflecting the views of the United States Department of Defense, spe- cifically, the US Navy or any of its component commands.
16 Acoustics Today • Spring 2021

   14   15   16   17   18