Had a look at doing this myself - looks like the main challenge is adapting your definition of words to split things like 22million.
At the moment I've got
def parse_alpha_numeric_word(char_block):
word_groups = re.split(r"(\d_)([a-z]_)", char_block)
return [word for word in word_groups if word != '']
Which does the job, but I can't figure out how to not match the empty strings either side (regex not my strong suit).

Then we have

split_words = number_sentence.split()
split_words = [word for char_block in split_words for word in parse_alpha_numeric_word(char_block)]

To get it back into a flat list of words

Thanks for pointing this issue @haydenc . Will look into this in the next version release. Also, you are most welcome to contribute and resolve this issue.

For now, added https://github.com/akshaynagpal/w2n/blob/master/word2number/w2n.py#L139 .