Giving DSi games their voice back

The DSi adds a small Teak coprocessor purely for audio: games hand it raw samples and it applies effects like reverb before the result reaches the speakers. melonDS’s DSi support had, until now, quietly skipped it. Games still played, just without whatever that chip was supposed to be doing to the sound.

Emulating the DSP properly, running its actual instruction set cycle for cycle, is a much bigger undertaking than the payoff justifies right now: it’s under-documented, and the games leaning on it are a small slice of the DSi library. So melonDS 1.1 takes the high-level route instead. Rather than executing the DSP’s program, it recognises the small set of known effect programs commercial games actually load onto it and reproduces their output directly.

That trade-off is exactly why microphone input landed in the same release. Both are DSi-side audio features that only became worth finishing once there was a decent baseline for the sound pipeline they run through to build on.

The honest caveat: high-level emulation only covers programs it recognises. A game running an unusual or hand-rolled DSP program will still be silent on that front. It’s a reasonable trade for now, most games use one of a handful of standard effect chains, but it isn’t the same guarantee the CPU and video emulation give you.