In the sense that like the video has its own RAM that just does video stuff then no.
There are aspects of the hardware that will quite happily use system memory.
Compared to some other things where you can pipe data directly from cartridges to the audio input buffer... not as much as some (the DS cart protocol is not something I would entirely want to rely on here but someone might have) but you could make it very tight if you wanted to, and can happily break things up to be sent in with very short bursts or otherwise play far larger songs than a simple memory copy might appear to allow if you had to have it all in RAM before starting.
The DS also supports PCM audio so if you can get it there fast enough then you are good.
If you are hacking games using the SDAT format (no data on the non SDAT stuff here though I would be surprised) then games not using the STRM (think stream, it mainly being PCM audio of longer form) sub format will not support such things as the compiler seems to strip it (got to save RAM after all) and people often then hack the game instead to use it in a slightly different way. https://nsmbhd.net/thread/3258-video-intro-with-streamed-sound/ being one example.
We have also seen people abuse the sound bank/sound font to stitch together essentially a wave audio track.