Friday 31 January 2025
For years, audio engineers have been trying to find a way to improve the sound quality of loudspeakers. One promising approach is called parametric array loudspeakers (PALs), which use ultrasound waves to create a highly directional beam of sound. However, PALs suffer from significant nonlinear distortions that can ruin the listening experience.
To combat these distortions, researchers have turned to deep learning, a type of artificial intelligence that can learn complex patterns in data. In a recent study, scientists used a type of neural network called WaveNet to identify and compensate for nonlinear distortions in PALs.
The team first trained their WaveNet model on a large dataset of audio signals and ultrasound waves generated by the PAL. The model learned to predict the output of the PAL based on its input and the properties of the ultrasound waves. Then, they used this model as an inverse filter to preprocess the input signal before it reaches the PAL.
The results were impressive: the deep learning-based approach significantly reduced both total harmonic distortion (THD) and intermodulation distortion (IMD), two common metrics for measuring nonlinear distortions. In fact, the THD was reduced by an average of 4.55%, while IMD was reduced by an average of 2.47%.
To put these numbers into perspective, traditional methods for compensating nonlinear distortions using Volterra filters can only reduce THD to around 5% and IMD to around 3%. The deep learning-based approach outperformed these methods by a significant margin.
The researchers also tested their approach on different frequencies, from 250 Hz to 8 kHz, which is the range where most music and speech sounds are concentrated. They found that the deep learning-based method performed well across all frequencies, reducing THD and IMD to below 5% and 3%, respectively.
This breakthrough has significant implications for the development of PALs and other audio systems. By using deep learning to identify and compensate for nonlinear distortions, researchers can create more accurate and faithful sound reproduction. This could lead to improved music and speech quality in a wide range of applications, from home entertainment systems to professional recording studios.
Overall, this study demonstrates the power of deep learning in improving the performance of PALs and other audio systems. By leveraging the complex patterns that deep learning can discover in data, researchers can create more accurate and efficient sound reproduction systems that benefit everyone.
Cite this article: “Deep Learning Enhances Sound Quality of Parametric Array Loudspeakers”, The Science Archive, 2025.
Parametric Array Loudspeakers, Deep Learning, Wavenet, Nonlinear Distortions, Audio Signals, Ultrasound Waves, Inverse Filter, Total Harmonic Distortion, Intermodulation Distortion, Volterra Filters
