Does anyone know how to stop sending agent speech transcription to the model? I get a lot of spelling issues because the model receives user + agent speech transcribed, instead of user speech + previous model output. here's an example User says "So please send the email directly to me at triston" Model outputs "I heard you say triston" Agent says this and it gets transcribed to "I heard you say tristin" (INCORRECT) Next OpenAI request messages array has "I heard you say tristin"