Training with data of 44.1khz

To train data with a sampling rate of 44.1 kHz, how should the configuration file or model be adjusted to avoid input-output length mismatch errors?
`[rank1]: RuntimeError: The size of tensor a (348) must match the size of tensor b (347) at non-singleton dimension 3
`