This feature consists in exposing to the Web the audio level of an encoded frame transmitted via RTCPeerConnection and exposed using WebRTC Encoded Transform.
audioLevel is exposed via other APIs (RTCStats, RTCContributingSources) and supports well-known use cases such as indicating who's talking in a VC application, or detecting silence. Having it as part of frame metadata makes audioLevel detection more accurate and efficient for applications using WebRTC Encode Transform, which work at the frame level. In this case, the application does not need to poll getStats() or getContributingSources() to get access to the audioLevel, and it will know that the audioLevel exactly corresponds to the frame being processed. It also unlocks other potential use cases, such as reducing the redundancy of frames with zero/low audio level.
Explainers: https://github.com/guidou/webrtc-encoded-transform/blob/master/audio_level.md