chunked/nltk_chunking/tasks_text-to-speech/chunk_103.txt

For example, try using config.reduction_factor = 1 to 
see if this improves the results.