Kickstarter

We only achieved about 19% of our goal pledges, but please consider donating directly to the project at http://talknicer.com/slics

While we are pursuing private royalty financing for development on the fragmented Android platforms, we could really use the help today to help collect exemplar pronunciations with Mechanical Turk from native English speakers with a wide variety of accents for as much initial instructional content as possible. We are offering essentially the same rewards listed at http://talknicer.com/slics#incentives -- please have a look, and please consider donating directly.

Thank you very much for your kind support and faith in the project team and goals.

Best regards,
James Salsman
jim@talknicer.com

The first code check-ins are available from launchpad (using the Bazaar source code control system) and subversion.  The current to-do list in on the project wiki, while the original milestones are on SourceForge.  The students are already weeks ahead of their original schedules.  

As always, check the project blog for at least weekly updates.  Both of this week's updates are posted.  Among many other things, Ronanki coded an edit distance scoring grammar generator based on neighboring phonemes, and Troy measured the automatic intelligibility versus bit rate bandwidth of the Speex vocoder at various quality levels.

The project's Google Summer of Code 2012 students' first weekly project reports were published today on the Pronunciation Evaluation blog. In them, Ronanki and Troy describe what they have done prior to the official start of their projects. Look at how long those lists are! They have already written several dozen kilobytes of code (I will let you know when they commit their first check in so you can follow along downloading it) including two different kinds of browser microphone upload applications (using wami-recorder and rtmplite.) Ronanki already has instant speech PCM upload with wami-recorder to sphinx3_align phoneme segmentation working, and a program which converts a string of phonemes into an edit distance scoring grammar for sphinx3. Troy has already measured the fidelity of the various speex vocoder quality levels for ultra low-bandwidth uploads to help save mobile battery life and network bandwidth. (I understand this might be very technical for most of you -- please ask me in comments here if you would like me to elaborate on any of these aspects after you read the students' blog posts.)

As they mention, both Ronanki and Troy will also be blogging weekly at the CMU Sphinx organization blog on SourceForge, which you might want to follow if you are interested in very technical aspects of speech recognition, and their project plans will be kept up to date on the CMU Sphinx SourceForge wiki. Troy also linked to his personal blog, where he has been keeping the detailed configuration and parameter information for all his projects.

In other news, O'Reilly Media featured us in the 4th position on their curated page linked from the Kickstarter home page. This should be great for our chances of success. As you can see, all of their featured projects have succeeded, by an average of 387%!

Here is our current Kicktraq progress graph:

It looks like we still need to work three times as hard, up from 3.2 times as hard a few days ago. Will do! (By the way, the success probability bump at $15,000 from the graph in my last update did turn out to be an artifact of insufficient data. Sorry about that.)

Finally, I have been informed by a potential backer that we need a $2,500 reward.  What should we offer? I'm going to start with having a biographical or any other suitable story selected by backer(s) at that level be included as one of the initial adventure stories we will ship with the default server and stand-alone instructional content. But I will also offer all other reasonable suggestions from existing backers posted in the comments here. I reserve the right to change the reward dollar amount in the case of particularly easy or difficult suggestions. So please start thinking -- what should we offer?

Thanks very much to the first ten backers! We have $440 pledged in just two days. That's more than I've raised for the project this year. I want to share this early update to show you Kicktraq, which can track the potential success of any Kickstarter project: http://www.kicktraq.com/projects/jsalsman/choose-your-reading-and-pronunciation-adventure 

Although at present Kicktraq shows the project meeting just under a third of its goal, do not dispair, because most successful projects tend to grow faster towards the end. But the reason they do so is because backers help, and I hope you will please share the project with your social media (including LinkedIn groups and status updates, especially, which are said to be better sources than Twitter, Facebook, and the like), on-topic mailing lists, friendly popular bloggers, and anyone you might know in the news media. It's a lot easier for backers to do a little more than for the project team to work 3.2 times as hard, but we will try our best to do as much as we can.

One more thing: Does anyone know if this success probability bump at $15,000 is real, or just an artifact of insufficient sample size? [Update: After examination of more data, it is now clear that it is just an artifact.]

(from http://edchedch.wordpress.com/2011/04/25/kickstarter-data-analysis-success-and-pricing near the end of the comments)

I didn't see that until after I picked the $15,000 goal, which was only based on our needs to get Guillem and basic content development funded. It's very encouraging, but it's also based on year-old data. I will do a screen-scrape of the most recent few thousand video game and software projects and see what they look like. If you know anyone thinking about choosing a Kickstarter goal, please blog about this update. Thank you all again.