chunked/nltk_chunking/tasks_text-to-speech/chunk_75.txt

Since the original 
target sequence length may be odd, the data collator makes sure to round the maximum length of the batch down to be a 
multiple of 2.
 

data_collator = TTSDataCollatorWithPadding(processor=processor)

Train the model
Load the pre-trained model from the same checkpoint as you used for loading the processor: 

from transformers import SpeechT5ForTextToSpeech
model = SpeechT5ForTextToSpeech.from_pretrained(checkpoint)

The use_cache=True option is incompatible with gradient checkpointing.