Tips For Best Results
Here are some recommendations that will help optimize the audio generated for your book and minimize errors.
Avoid hyphens.
AI often sees hyphens as pauses. So a word like “bird-like” will sometimes be generated as “bird” and “like” with a slight pause in between, as if it were actually an em dash. This is inconsistent and depends on the words linked by a hyphen, but in general it’s best to avoid them. A simple way to prep your manuscript is to do a find and replace, with the hyphen replaced with a space.
Separate acronym letters with spaces.
While we would speak the acronym VIP as three different letters spoken, AI will sometimes not recognize them as an acronym and will pronounce them like “vipp.” This won’t happen if the letters are separated with a space like “V I P,” which will lead to the acronym being spoken correctly.
For unique names and words with difficult pronunciation, spell them phonetically.
This is especially important for fantasy authors with unique names that have odd spellings. AI will often have trouble pronouncing a name like “Mistyre.” If you hear it as pronounced “mistear” the AI may pronounce it as “mistire.” It is difficult to anticipate what words generative AI will have trouble with, so its best to err on the side of caution. These are generally not too difficult to prep, as find and replace does the trick.
Replace underlines and italics with all caps.
Generative AI tends to ignore word styles, and they won’t affect how the words are read. However, words in all caps will often lead to the voice lending more emphasis to the reading. So it is recommended to replace any italics or underlines that you use for emphasis with all capital letters.
Split long paragraphs into shorter ones.
Generative AI often has trouble with longer audio generations, with the latter part not sounding as good or as consistent as the beginning. One way around this is to split your paragraphs into smaller ones. It really doesn’t matter where, as most audiobook pauses between sentences aren’t too different than the pauses between paragraphs.