It is a well-known phenomenon that a DAC typically plays on for a while after the network cable is disconnected, due to buffering. I have recently been beta-testing a new DAC with AI-supported buffering (sorry, I can’t disclose the brand yet). This technology continuously monitors the music and detects patterns in both harmonics and lyrics, which enables it to extrapolate the music even after the network buffer is exhausted.
How well this works depends on the predictability of the music. With Bach, today’s algorithm can extrapolate several seconds, with Stravinsky half a second. With typical pop music today, the AI can continue playing plausible music about 30 seconds. With 1970s disco it plays indefinitely without network connection. The AI still has not managed to synthesize avant-garde jazz.
Research continues. Machine learning based on the vast library of music collected in the internet is rapidly extending the extrapolation algorithm beyond current capabilities. It is of course a race between machine learning based on music archives and composition of more complex music. Many music theorists expect, based on current musical styles, that machine learning will rapidly outpace current composition and will soon be able to play most modern music indefinitely without a network feed, eliminating the network challenges we face.
<April 1 is coming up…