hifigan lite model on TensorFlowLiteSwift error: input_channels % filter_input_channels != 0 (12!=0)

[lalalic]

I'm working with fastspeech2_quant.tflite and hifigan_float16.tflite on TensorFlowLiteSwift

The fastspeech output tensor's shape is [1,92,80], and there's an error when resizing input tensor of hifi-gan model with shape [1, 92, 80]: input_channels % filter_input_channels != 0 (12!=0).

I tried to extend the data to [1, 160, 80], and then there's no such error to resize and infer, but the output from hifi-gan is just noise.

Any idea?