How realistic do AI narrators sound?
Dreamstime (TNS)
” Human readers, unite!” wrote a New Brighton, Minnesota, reader.
Ever since I wrote a column about Google and other platforms using artificial intelligence (AI) voices for audiobooks, my inbox has been slammed with spammy emails about AI.
“An AI Bot Wrote This Copy,” reads one message line. “Influencers helped by artificial intelligence,” reads another. “Can AI Teach Us to Read?” reads yet another. And then there’s “Why the fuss over ChatGPT?” Oh, man, I just cannot get into ChatGPT at this stage of my life.
Much better were the many emails and comments sent by actual humans. Thank you!
Most of you found the concept of using AI readers for audiobooks to be abysmal.
” Human readers, unite!” wrote Stan Kaufman of New Brighton, Minnesota. “I would prefer a talking book written by, say, Edgar Allan Poe and read by, say, Basil Rathbone, over the same book read by an AI algorithm. The reader’s name should be a prominent part of the package. Can one’s voice be copyrighted?”
One’s voice cannot be copyrighted, but there are laws popping up to protect humans in other ways. In February, Vice magazine reported that voice actors are being asked to sign away their rights to their own voices. In an earlier story it reported that the voices of celebrities — Joe Rogan and Emma Watson, to name two — are being “cloned” and made to say racist things.
Yikes! But I digress.
Readers wrote to say that they want to listen to human readers. For one thing, they have concerns about a robotic voice getting the tone right.
“I began Talking Book [audiobooks for people with poor vision] after I was diagnosed with macular degeneration in 1956,” writes Ron Weitbrecht of Cambridge, Minnesota. “I cannot imagine how awful it would be to hear the type of robotic voices I use on my computer or my GPS as I try to plow through a great book!
“The human voices involved in such endeavors must be well-trained and practiced at the art of storytelling. I for one do not wish AI to insert itself into my Talking Books.”
Patricia Kovel-Jarboe has been a reader for Minnesota Talking Book for many years, and she, too, wonders about tone. “It’s hard to imagine AI being sophisticated enough to know when a character in a novel is using a ‘snarky’ tone of voice or that a teenaged character is using that tone that comes with a built-in eye roll,” she writes.
Jerry Mandaville of Maple Grove, Minnesota, writes that “being legally blind, listening to books fills an important part of my day. I am at the point now where the narrator is as important as the book. In fact, I will seek narrators as I decide on what book to listen to next.”
I must play devil’s advocate here; I prefer real people to bots too, but it should be noted that AI voices are getting better — more realistic, less like those weird atonal voices we first encountered.
As Laura Miller writes in Slate, some of them are actually pretty good, though she also notes they have a ways to go.
Beth Sartor Obermeyer says that when her latest book was being made into an audiobook, “the researcher for the audiobook company called me to confirm the pronunciation of 20 pages of words in the manuscript. He even called the library in tiny Titonka, Iowa, to get the right twist on the names of generations past. I am so proud of the effort.”
An effort that a bot, almost certainly, would not make.
