Upvoted for usefulness although I wish you'd resisted the temptation to pick a cute title and gone with, "The sequences in MP3 format."
Good point. changed.
Now I'm curious to know the original title. :D
AT&T Mike meets LessWrong.com
I have created a full collection of all the sequences in audio format. I looked in to releasing it as a podcast but discovered that it would not be legal (without paying the appropriate fees).
It may actually be worthwhile for the SIAI to buy a T2S licence and add a podcast form to LW.
LessWrong might be exempt from these fees, but the exemption is unclear. Specifically, I don't know how to interpret "other consideration of any kind". I am not a lawyer.
From the license holder's website:
5) Do I need a license to distribute mp3 or mp3surround encoded content?
Yes. A license is needed for commercial (i.e., revenue-generating) use of mp3/mp3PRO in broadcast systems (terrestrial, satellite, cable and/or other distribution channels), streaming applications (via Internet, intranets and/or other networks), other content distribution systems (pay-audio or audio-on-demand applications and the like) or for use of mp3/mp3PRO on physical media (compact discs, digital versatile discs, semiconductor chips, hard drives, memory cards and the like).
However, no license is needed for private, non-commercial activities (e.g., home-entertainment, receiving broadcasts and creating a personal music library), not generating revenue or other consideration of any kind or for entities with associated annual gross revenue less than US$ 100 000.00.
Oh. I understand now. I've tried a few text-to-speech engines and AT&T Natural Voices sound the best to my ears. I will find the terms of use and pricing for that.
Please do. If they are sufficiently cheap I will see about getting someone here to allow me to implement an automatic audio version of either just the early Eliezer posts (including sequences) or as a feature of all posts. This would be massively valuable for many of us.
In fact, if price is prohibitive I wonder if it would be worth implementing a free (less natural sounding) text-to-speech converter.
Other questions to resolve:
Where should the files be hosted? (Does LW have the bandwidth) Probably, it isn't a huge amount. If not I have half a dozen servers floating around the place. They cost pittance.
Is LW exempt from MP3 licensing? (I hope so)
Probably but I know less than you.
Where should the download links be placed? (A wiki page is fine, but it will be less discoverable.)
A wiki page sounds good for now. If people find it especially useful we can work from there. (I may create an RSS feed or podcast at some stage if I feel inspired.)
Which posts should be completed first?
Whatever you happen to care about.
I have price quotes for Acapela, Cepstral, Wizzard (AT&T Voices), Neospeech, and Nuance RealSpeak. The range is from $1,000 to $15,000 USD.
Open source options are eSpeak (robotic), Festival (robotic), FreeTTS (robotic), Pico and others.
Pico is part of Android and it sounds more natural than other open source options I tried. Pico is licensed under Apache 2.0. Here's a demo.
The commercial voices are definately better; Loquendo is a good example.
So now I can start converting via Pico or try to get funding for a more natural voice. Thoughts?
Converting is easy, but what's the next best format?
WAV is going to need a lot of bandwidth and storage. Ogg Vorbis works fine on Android, but not for the iPhone.
MP3 will give the best user experience for the most users.
This seems like a pretty good idea. You should make a wiki article with links to different sequences.
Great idea, but don't forget about copyrights.
I can drive and listen, but I can't drive and read! The same is true for most kinds of exercise.
If you are in my situation - wanting to read the sequences without having enough time - feel free to download these audio files for your smart phone or MP3 player.
My vision is to build a podcast feed or downloadable MP3 repository of all the major sequences. The files I have now are not organized enough to scale out to hundreds of posts, and some of the artifacts of text-to-speech could be reduced with the right pre-processing. Before I spend more time on this, I want the right tools and process in place.
Any ideas on how to to proceed? Would you like to help? How should I publish these files?