Change sample rate for pre-trained BrowserFft model

I am new to tensorflow, when I follow speech recognition model_maker tutorial , found the audio sample rate is fixed to 44100Hz, but I want to use 16000Hz for real-time command recognition on my poor Pi zero 2W. Although I can resample audio, but in my test, resample process always cost about 600ms, which is not good for real-time purpose.
I also tried modify sample rate in class BrowserFFTSpec( :

  _SAMPLE_RATE = 16000
  def target_sample_rate(self):
    return 16000

and get error info “ValueError: Input 0 of layer “audio_preproc” is incompatible with the layer: expected shape=(None, 44032), found shape=(1, 16000)”. So I wander if there is a setting in “tfjs-sc-model” model files, but no idea what to do next.

Any help would be greatly appreciated!