Wav2vec2 : expected sequence of length 49863 at dim 2 (got 68198)

I am getting below error while fine tuning facebook/wav2vec2-base on custom data

expected sequence of length 49863 at dim 2 (got 68198)

in my dataset following text has audio file with input values length 68198
the infundibulum can be seen if it is floating in peritoneal fluid

Also following text has audio file length 49863
overall coarse and heterogeneous echotexture

I am guessing these two strings might be getting in the same batch and that might be causing the problem.

tokenizer = Wav2Vec2CTCTokenizer("./vocab.json", unk_token="[UNK]",
pad_token="[PAD]", word_delimiter_token="|",
padding = True)
also tried padding = “max_length” , but still got the same error

data_collator = DataCollatorCTCWithPadding(processor=processor, padding=True)

How to fix this?