I gather injecting audio is "forbidden" by Google, microphone goes codec > baseband but apparently there are ways at the kernel level to get around this, Reasons for this ? To auto play "No consent is given to record this call" or recording "pips" during call to keep them honest, or "My audio is copyrighted material that may not be stored or used without written authority and wet ink signature" or "All calls will be recorded to catch marketing misrepresentations or promises for enforcement", auto legal disclaimers (like they cram down our throats) and such. Level the playing field. Secondly for injecting audio for the many scammers, like chimpanzees screeching, toilet flushing, 1970's porno soundtracks, etc, maybe even dump them into a Kafkaesque voice tree app like they dish out to us. Auto-play audio scripts also has uses in regulated industries, and you have almost certainly had it used on you. Well, turn the tables !
There is a serious side to this, with "them" having a robot saying "you consent to X" and you having a one touch robot voice to say otherwise that will be a big deal with their trying to steal biometric data without consent bya "we will do X", where your silence becomes consent unless you keep objecting. Claiming copyright of your talk is especially potent and setting them up and they cannot easily get out of enforcement, especially where joint copyright is claimed and they need the consent of ALL the joint holders.
Anyone up for a challenge ? Our company will donate to graphene and/or pay someone to make this happen.