This is done entirely on the streamed sound data, unlike the old relative volume which uses the backend's volume setting.