A statistical examination of the predictive audio buffering algorhythm in case of musical content
A lecture by Gabor Gerenyi, Overmind Ltd.
Abstract
Introduction
Multichannel audio streaming is a widely used technology in today’s digital audio workstations. Since high-resolution uncompressed audio streams put a high demand on both the hardware systems and the network infrastructure it is a necessity to use optimal buffering strategies. To monitor its efficiency we used a heuristic approach and a series of closed-loop experiences for our research set to recover possible weak points and establishing some conclusions.
Materials and methods
Experiments were performed on a moderate best-buy-for-studios Windows 10 system featuring a 8th generation Intel Core i5 microprocessor (3 GHz clock frequency), 16 GB of RAM memory, Western Digital Elements 1 TB external storage for caching and buffering (USB3) and a 100 megabit per secundum internet access. Bandwidth is validated by Ookla speedtest before every experiment and Google Chrome’s Throttling function set the actual exact bandwidth as desired.
Server-side bandwidth is fairly all above (more than ten times) the necessary amount.
Results
Statistical approach showed digital audio workstation hardware systems can perform far better in streaming multichannel audio via public internet infrastructure than the calculated maximum performance due to the nature of the actual music structures. When not using any concurent application opened, playback a typical projekt size of 30 high-resolution stereo tracks is managaeble with the test configuration using Google Chrome browser.
While we measured the system’s performance as a whole and not just the elements – actaul performance is a function of several other elements influencing it like operating system’s service function, browser and sound card delay, Webaudio API etc. - the experiments prove they can use by our statistical methods and in normal conditions (no errors, normal temperature etc.) they produce very consistent and predictable performance data.
Conclusion
Digital audio workstations using streamed audio data from the cloud can be soon the mainstream approach in the near future - technically it is already a viable option, and statistical research and smart predictive caching algorhythms are still quickly developing areas. Finding new use cases, new usage models can unlock the existing technical potential and gain new creative audiences.
