Discrepancy regarding DecodeMsPNS_S32_I and MDCTInv_S32_S16
According to OpenMax DL specification, MDCT_Inv_S32_S16 provides output in Q15 format.
As ISO/IEC 14496-3 specification suggests that we should use Q15 format for input data while performing perceptual noise subsitution.
But, DecodeMsPNS_S32_I takes input as in Q28.3 format, which requires conversion of Q15 to Q28.3, as we have to take output of previous frame as input to this function.
Wouldn't it be more appropriate, if DecodeMsPNS interface also took Q15 data as input, so as to avoid this conversion?
Is there some specific reason behind it?