That's a pretty narrow view of "accessibility". For example, you just assumed that your user can't see (or can't see very well?) but can hear.
Users who are deaf and blind? Out of luck. Users who are deaf and not blind but need your thing zoomed to a larger size? Maybe out of luck maybe not (depends on whether the WebGL app detects browser zoom and actively works to defeat it like some do). Users who are deaf and not particularly blind but happen to not be able to tell apart the colors you chose to use in your WebGL? Also out of luck.
What the DOM gives you is a semantic representation that the user can then have their user agent present to them in a way that works best for them. Reproducing that on top of WebGL or canvas really is quite a bit of effort if you really want to target all users and not just a favored few groups.
Users who are deaf and blind? Out of luck. Users who are deaf and not blind but need your thing zoomed to a larger size? Maybe out of luck maybe not (depends on whether the WebGL app detects browser zoom and actively works to defeat it like some do). Users who are deaf and not particularly blind but happen to not be able to tell apart the colors you chose to use in your WebGL? Also out of luck.
What the DOM gives you is a semantic representation that the user can then have their user agent present to them in a way that works best for them. Reproducing that on top of WebGL or canvas really is quite a bit of effort if you really want to target all users and not just a favored few groups.