> the following audio processing is better done after converting the samples to FP32

Or, in some cases, FP64.