My bet would be 'reading'. Reading is essentially a type of synesthesia that we force onto children which turns external symbols into sound (at least initially). My interpretation of Jaynes is that he posits a direct 'voice' from the fast pattern matching systems of the brain (this, in turn possibly a maladaptation to the emergence of language earlier in our rescent evolution). Perhaps reading disrupts this pathway or teaches us to regulate it much more tightly? A prediction of this might be that voice hearing could be far more prevelant in illiterate people that we suspect.