Sounds are compiled by the native sound compiler, but we get first look at the compile here. We read the lipsync visemes and subtitles authored in the sound editor from the sound's .meta and inject them into the compiled resource as custom blocks, then return false so the native compiler runs as usual - our blocks ride along into the final file. The engine reads them back out via Sandbox.SoundFile.Visemes and Sandbox.SoundFile.Subtitles.