November 17, 2012

Status update

I wish that this were a progress report instead of a status update, but so far we haven't raised enough to begin data collection with Mechanical Turk.  We have had a paper accepted for publication and we are trying to get in to the Google Compute Engine to save expenses for the huge Amazon bill for asking people who claim to have good pronunciation and reading skill to record exemplars. The problem is that the number of such exemplars needs to be relatively large. For those of you familiar with the TalkNicer demo, this is the "exemplar sufficiency index" and it needs to meet a certain threshold for at least 5,000 words of instructional material before I feel comfortable committing to an expensive data collection effort.

So in summary, please donate more, or if you have already donated, please ask multiple people to at least match your donation. It will be worth it.

Update: How much more do we need? About $4,000 based on the preliminary per-phoneme exemplar sufficiency index including English homographs and Mechanical Turk performance expectation estimates. Also updated:

June 3, 2012

Weekly Updates on the GSoC Project Blog

Check the Google Summer of Code project blog for at least weekly updates. Both of this week's updates from the two GSoC students 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 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.

May 29, 2012


I just received this in email:

"While your motives may be pure, I could lose what little consulting work I have if this gets produced--and since my husband (a journalist) has been out of work for 4 years, and I make less at my full-time teaching job than I did right out of college 20 years ago--I beg you to reconsider offering products for free. It doesn't help people, it only hurts them. Without my consulting income, we run out of money by the 15th every month."

I've never felt so conflicted in my life. Comments on this are most welcome.

April 25, 2012

Google Summer of Code is on!

Google's 2012 Summer of Code has accepted two students to work on a pronunciation evaluation system for language learning over the summer. I'm honored to be mentoring Srikanth Ronanki and Li Bo (Troy) for the CMU Sphinx organization, which has six students this year. It's my third year as a mentor, and the second year for Sphinx, so I'm pretty optimistic. This will get a robust open source pronunciation assessment system for language learning out to the widest possible user base in the most efficient way imaginable. You can read all about the application process, the students, and the text of their applications at

February 29, 2012

Recent MIT Ph.D. thesis on crowdsourced pronunciation assessment

I've received a handful of emails asking where to read about the latest technology in pronunciation assessment for language learning. While there's plenty to see in Google Scholar, too much of it is overly specialized or otherwise esoteric. Mitchell Peabody earned his Ph.D. from MIT last year and published this fantastic thesis which not only covers all the important aspects of the field (including both acoustic and edit distance scoring hybridization) but does so in a way which should be accessible for most people with a basic computer science background. But the best part is that it goes into great detail on crowdsourced phonetic labeling, as well as what Peabody calls "vowel anchoring" which I think is a tremendous improvement in scoring. Add it to your reading list!

January 29, 2012

Ask questions here

Use the comments on this post to ask questions which aren't already in the Q & A. I'll answer them in subsequent blog posts. Include your email, subscribe to RSS or follow to make sure you see the reply. Comments are moderator so spammers won't see your address.

January 28, 2012

Android client demo

Here's a demo of the basic client-server demo on an Android phone: