Page 1 of 1

Human Read vs Computer Generated (TTS) Audiobooks

Posted: January 5th, 2021, 9:45 am
by MaryAnnSpiegel
As we begin a New Year, I’ve been thinking about the unique place that LibriVox fills in an increasingly technological world. Our FAQs put it this way:
hugh wrote:
January 10th, 2006, 8:47 am
What's this all about then?
At LibriVox, volunteers record chapters of books in the public domain into digital audio (eg. mp3), and then make the audio files available to the world, for free (through our catalog, a podcast, and in a variety of other formats at Archive.org)

Why are you doing this? What's in it for you?
We love reading, love books, love literature, think the public domain should be defended and enriched, we like free stuff, we like to hear people read to us, and we like reading to other people. It's fun, it's a great community, it's a rewarding public service to the world. And "nothing" is in it for us, except the satisfaction of participating in a wonderful project.
The LibriVox project and community exists to create and share human read audiobooks. By that I mean audio recordings of books being read aloud in real time by humans.

Is there any other kind of audiobook? Well, yes. Text-to-speech (TTS) technology speaks aloud digital text using a computer-generated voice, or a highly manipulated human voice. There are plenty of websites and software programs out there where you can convert text to an MP3 file. There’s a lot of money to be made commercially by having TTS systems sound like real people. Whether it’s Google, DeepMind, WaveNet, or a myriad of others, the technology is advancing quickly. I respect the people who are developing these systems, and those who would like to use these systems to improve the lives of others. I’m sure the day is coming when a multitude of TTS audiobooks will be available, on both subscription and free platforms.

But LibriVox is not a platform for creating and sharing TTS audiobooks, it’s a place for real people to read to real people. We value the act of reading, we love communicating with our listeners, adding our own interpretation, friendliness and expression to all that we create. We've worked hard to make LibriVox a welcoming place for our volunteers to share their love of reading aloud with others.

So why do I bring this up now? Well, because last year we had several volunteers (BC/DPLs) flag files as questionable. We ended up rejecting several files after determining that they were not read in real time by a real person. We've also had potential volunteers tell us that they would like to use TTS technology in creating files for LibriVox and we've had to explain to them that LibriVox is a site for audiobooks read in real time by real people.

If you’d like to experiment with TTS, wonderful. If you’d like, you can use TTS to "read" a whole book and upload it to Archive.

But please, don’t submit a TTS file to LibriVox. Don’t try to pass it off as a real person reading out loud in real time. That’s not what we are about. Please respect LibriVox and our community and keep your readings real.

Thanks to all of the volunteers who made 2020 a record breaking year for LibriVox and for free [human-read] audiobooks! I love working with you and look forward to another banner year in 2021!

And if you are a BC or a DPL who suspects that a voice may be artificial, please send a PM to your MC and let them know. The admins will review the file and decide how best to address the situation.

MaryAnn

Re: Human Read vs Computer Generated (TTS) Audiobooks

Posted: January 24th, 2021, 2:19 pm
by TheDinosaurPlanet
I appreciate the fact that Librivox only includes recordings that feature real people reading or voicing lines.