SPEECH DELAY, Length of created audio from SSML doesnt match my SSML file

Hello, i am using azure congnite services API where i try to transform SSML into audio which work well expect my sentences doesnt start at the right time for example my last sentence "Do hrnca dáme bravčovú masť?" should start at 35290ms but starts much later about 40 second and this goes for every sentence and that results that my final audio is much longer.