The W3s WCAG Accessibility Guidelines:
basically states to remember that SOME users are using screen-readers (so you want to avoid auto-start any contrasting audio), and of course others are deaf (so whatever is supposed to be communicated via audio needs to be communicated another way).
Some of the "sufficient techniques" listed include:
- G60: Playing a sound that turns off automatically within three seconds
- G170: Providing a control near the beginning of the Web page that turns off sounds that play automatically
- G171: Playing sounds only on user request
- FLASH34: Turning off sounds that play automatically when an assistive technology is detected (Flash)
Their example for visuals is, instead of a green arrow at the bottom of a page to indicate "continue", have both the alt text AND text in the graphic say "press this arrow to continue" (and perhaps add a black outline, so color-blind people can see the contrast more easily.)
For audio, I think that instead of ONLY relying on a sound to indicate "success" or "fail", think of it as "frosting" - just a reinforcement of the basic text indication.
NNGroup hasn't done a ton on audio recently, but they have been looking at Audio-Interfaces (such as Siri, Alexa, HeyGoogle, and more), and it divides audio cues into a few kinds:
Nonverbal sounds, or earcons (auditory icons), which are distinctive noises generated by the system, usually associated with specific actions or states. For example, Siri emits a 2-tone beep after detecting its activation phrase, to signal that it is now ‘listening’ for a command.
Explicit verbal signifiers, when the system verbalizes a suggestion or request to let the user know what commands are available. For example, if you tell Google Home to “Set a timer,” it responds with “Ok, for how long?”
Implicit verbal cues, when the system hints that an action is possible, without fully articulating the suggestion.
The article goes into more detail, indicating how just as visual icons can be confusing, the same goes even more so with Earcons and implicit cues. I don't know if this audio-only interface information is what you're looking for, but I thought it may help.
Update July 2019:
Google has added sound to their Material foundation.