Annals of Wu

a sinotibetoburman linguistics blog

Phonemica 4.0 Changelog phonemica - development

In keeping with the theme of "Annals of Wu as Phonemica devblog", I thought I'd outline the changes in the latest rewriting of the site. I spent a week of 14 hour days this past week rewriting the whole system, almost from scractch. Here's what's new.

version 3.9 changes

These are old changes from a few weeks ago that mostly just deal with the transcript and segmenting, but they're part of the larger push to give people more power and offer greater automation.

  • users can now control segmenting of audio better
    • you can drag segments to other timings
    • added ability to add segments by clicking anywhere in the waveform
    • added said waveform to help visualise the recording for segmenting
  • vastly improved automatic scrolling
  • included a second copy of the transcript in plain text further down, by request

  • version 4.0 changes

  • greater control for editing information on your uploaded storytellers
  • can now upload multiple recordings under a single speaker
  • specify if a story is told in a dialect other than the speaker's native dialect
  • upload storyteller photos directly to the site, available immediately
  • upload a wider range of audio formats to the site
  • a Korean version of the UI is now available, though it needs a few more translations. thanks go to Minkyu for helping get that started
  • profile pages are back. show off what you've done on the site.
  • delete stories you've uploaded, like for example if you upload the same one twice. that's happened a couple times too

  • Most importly, as of this latest update:
  • your submitted audio is available on the site immediately.

  • There's no more need to wait for us to hand-process every submission. Up until now, due to limitations with our old hosting company, we simply couldn't automate this in a way that was satisfactory to us.

    Other less visible changes

    First, we've changed hosting companies. The best part about this change is that we're off shared hosting and on a VPS, which means we now have full control of the server that the site is on. This let me upload all the software and write all the code I needed to to allow for the above automation. This flexibility is also giving us even greater opportunities for future improvements already in the works.

    Second, the site has been moved to a server based in Tokyo rather than Singapore. This has already proven to be a great benefit for site speeds for users in China, which is most of our users at this point.

    The database has been completely redone. We've moved a lot of the data off MySQL and onto NoSQL. Without getting too much into it, this allows us to put less strain on the server in peak use periods, and also lets us organise our data a bit better. The old system was a bunch of revisions and amendments and changes built on top of each other going back to 2011. It was clunky and we hated it. It lead to a couple site crashes and slow load speeds.

    We've also done a lot of opimisation. I'm a reasonably capable programmer. I wasn't when we started. There was a lot of bad legacy code that could have been improved on. We didn't do that, opting instead to throw it out and start fresh.

    Changes coming this weekend

  • reassign stories from one speaker to another in case you upload it for the wrong person. it's happened
  • customise your profile by adding a picture and specifying your mother tongue, and whatever else you want people to know

  • Here's the thing. We're trusting you, the user, to take care of your own recordings and the associated data. But then in the past version, we didn't give you enough control of that. You still had to rely on us, usually really just me, to help you out with that. But I don't have that sort of bandwidth in my life. This latest round of changes is meant to kill two birds with one stone. You get more control, and I have some time freed up to work on more improvements rather than going around and fixing earlier slipups.

    We think you'll like the new site. We do. There are still a couple bugs to work out, and some features to implement. But feel free to comment here with feedback/bugs anwyay. Or just comments in general.

    I'll update this post when the "4.1" changes get made in a couple days.

      An Move To Better Transcriptions And Offline Support phonemica - development

      We're at a point in our lives where we need to make some hard decisions about Phonemica. It's not sustainable to keep it going the way it is. Expenses are out of pocket, and we're just not able to keep funneling earnings from our day jobs into the project when there are so many other bills that also need paying. Since we're not ready to give up on the project (and I probably won't ever be at that point), one of the things that absolutely has to happen with the project is better automation. We need to be able to give users more control over the data, and make things so ridiculously easy to use that our miniscule but awesome team doesn't need to be in there manually processing everything. A huge step toward that end is the new transcript system, which I've recent about previously.

      Another important step is to get more linguists involved. Not necessarily in consultant roles or active with other people's recordings. Instead, we'd like to see more people using Phonemica as a platform for sharing their data with the world. While we understand the desire to keep things under wraps while in the publishing process, we're strong believers in open data and making things available. We would also love to see people fine tuning transcriptions and improving on the data.

      One way we're doing that is by offering exports of the transcript data for use with software like EXMARaLDA. Starting now, transcripts from entries can be downloaded as .eaf transcript files for use with ELAN. These are generated on the fly, so the file will reflect the transcript as it is the moment you download it.

      In addition to this, I'm nearly done coding an .eaf importer. It will be made available for people who wish to work on transcripts on their computers and then upload the files to Phonemica to be converted into an online entry. To better integrate this, we need to establish better entry ownership so that people can lock and hide their own submissions while they work on the data offline. In general this is somethign we want to do anyway just to give people more control, and so that they can prepare the entry to completion before letting it be public.

      The new transcript system is going up tonight. The .eaf downloads are already there, but a little bit hidden at the bottom of the entry pages. With a little luck the .eaf importer will be done some time next week.

      These are all thigns we wanted to have from the start but never had the resources. Now that the site is up and stable and has the data and users to test new features with, we can start to integrate these ideas that until now had never seen the light of day. There's much more we want to include in the future, and we will keep trying to get that up.

      In the mean time, we think these last few changes will really impact the way you use the site in a very positive way.

        Displaying Tone Contours On-Site phonemica - development

        I've already mentioned the new transcript system and segment editor. There's actually one more big thing that we're adding to that. The waveform is new to most users, though we've had it implemented for site admins for some time now. However the new version is much less demanding on the server or users' download times.

        But in addition to the waveform we're adding something else, seen here in red:

        While not present on all entries, many will start to get pitch contours added to the display. Does this help the transcription? No not really. Does it do anything for the average user? Probably not, unless they're uncertain of what tone is being produced and this somehow clears it up for them. So why do it?

        One of the big things we set out to be from the very beginning was an openly accesisble source for data on Sinotibetan dialects. It was our hope that this would eventually become a resource for academics who are working to better document the language family. Over the past few years that we've developed the system, that has never changed. The goal will always be to increase the quality and amount of data available for analysis.

        One big area of analysis in Sinotibetan is tones. There's a lot of good work that's been done in the past, and a lot of serious data gathering and analysis that is still ongoing today. By including tonal contours in the data on Phonemica, we're hoping it can be one more reason way to quickly communicate the data to researchers without them having to go through the extra steps to display the contours themselves.

        Of course, they could. Anyone could open up the audio in Praat and quickly get a contour. For the point of analysis where frequencies are being analyses, that will still probably be the way to go. But for the inital inspection, presenting the data in a quickly visible way may prove useful for some people.

        There are other reasons to include it as well, such as simply wanting to develop better tools for documentation. I'll get into that in a little more detail in a later post.

          Improved Segmenting On Phonemica phonemica - development

          Segmenting is a pain. It's the most time consuming part of processing recordings for the site.

          In general, for each file that's submitted, about 30 minutes gets spent in preparing the story for transcription. About 10 minutes of that is spent on file preparation and then 5 on getting the database information corrected and another 15 for cutting the sound up into the segments you see in the transcript.

          It's just not sustainable. I'm the only one doing it, and it's incredibly time consuming when the submissions start building up. To solve the processing time before segmenting, I've written a script that runs on the main Phonemica computer which automates all the file conversion work that needs to be done. There's a lot of it.

          The other major improvement is with the segmenter. The site administrators have access to a tool that allows them to segment recordings. This involves working with a waveform that goes along with the audio, and then manually marking the start and stop of each segment. It works, and it's what we've used for the past year, but it's pretty awful for how much time it still takes to do the work.

          Starting this coming weekend, that iteration of the tool will be deleted forever. In it's place, we now have something much more useful. Below is a shot of the new system.

          This is the new version. Note that this isn't just a replacement for the segmenter tool that you've probably never seen or heard of before now; It's also a replacement for the old transcript system on the pages of individual entries.

          For users, this will be something immediately available for transcription and cleanup. For me, the bigger part is replacing the segmenter, since that was the major time suck on our end. Rather than me having to go through and segment, and then only other admins being able to edit, now anyone who's signed in can not only edit the transcription as before, but also fine tune the timing of each phrase. If you don't like how an entry is segmented, and there are plenty that need cleaning up, you now have the power to do something about it. Want to add another segment? Go for it. Are there too many segments in a small space? Now you can delete them. Don't delete all of them though, becuase that would stuck.

          We really think this is going to be a huge improvement in usability, and we hope you'll agree. This is part of a larger push to give people more control over their stories, as well as getting the data in better shape for future uses by researchers or language learners.

          If you have any questions, feel free to shoot me an email. Otherwise stay tuned as the update should be in place by the weekend.


            A semi-academic linguistics blog about Sinotibetan, previously focused primarily on Wú, a Sinitic language spoken in the Yangtze Delta region. Topics now include historical linguistics, documentation, language rights, sociolinguistics and learning materials, as well as acting as the dev blog for Phonemica from time to time.

            I'm a linguist based in Asia, working on documentation and historical development of Sinotibetan. In addition to academic research, I'm heavily involved in Phonemica, an organisation that promotes crowd-sourced preservation of local languages.

            I'm currently in the field, so getting in touch isn't easy. However you can try to email me at the following address and I'll respond as soon as I'm able:

            © 2009-2017