Dear friends of DARLA,
(1) Our DARLA system ran into a technical bump in mid-July (July 16, 2026). The DARLA system is hosted here at Dartmouth (since 2015). The Dartmouth IT security team notified us that DARLA can't be a public-facing tool anymore. There has been no security breach in DARLA. However, DARLA's code is 11 years old, and the world out there has lots of new AI hacking risks, etc. Since DARLA receives online uploads from users via a public platform, security is an institutional concern. We apologize for the inconvenience, but we just have no control over these policies. Please contact me at James.N.Stanford@Dartmouth.edu if you need more information about any of this.

***Yes, we will be able to continue to process your jobs with DARLA. It's just going to require an extra step: We are offering a human step to make it possible to still do your DARLA jobs: Specifically, we have hired a Dartmouth student RA to manage uploading your files by hand. Please see (3) below for the details.

(2) But for the longer term, please note that DARLA will be permanently riding off into the sunset at end of January 2027. The reason is that the technology is quite old. My postdoc Sravana Reddy built DARLA in 2015 -- that's an eternity in tech years -- there are now much newer, better systems available for vowel extraction. In particular, we highly recommend "New FAVE" (designed by Joe Fruehwald), which has better vowel formant extractions than the system we have in DARLA (which uses "old FAVE Extract" from 2015). Both systems currently use Montreal Forced Aligner for alignment, which is state-of-the-art, but the vowel formant extraction system is much better in New FAVE than in DARLA. Currently, there is no user-friendly system like DARLA available to run jobs in New FAVE, and we won't be able to put it into DARLA. However, help is on the way at another school! Joey Stanley (BYU) is finishing up a brand-new system, a user-friendly interface for New FAVE, which he built. This new system, VoxHumana, will replace DARLA very effectively! VoxHumana should be available toward the end of this summer, based at BYU. VoxHumana will be a user-friendly online uploading system much like DARLA, where you can upload audio files and get alignments and vowel formant extractions. It will also have Whisper auto-transcriptions, and also it will have additional post-processing steps like normalization options and vowel plotting options that will be very handy. VoxHumana is currently targeted to be ready for the end of summer (see Joey Stanley at BYU for more details).

(3) For now with DARLA, here are the steps you can take to submit a job: We have hired a Dartmouth student RA to manage uploading the files by hand. The RA will simply upload your job for you using the DARLA system as normal from our campus system, and then DARLA will directly send the output results to you as normal. We will maintain data confidentiality as normal. The only difference is that a human RA will be the one uploading your files into the DARLA system.

Send your job to the Dartmouth student RA email address here: beepboopbeepboop28@gmail.com
That mailbox will be checked daily.

Please send as one email per job so that it will be easy to organize in the RA's inbox. In the email, attach your transcription (TextGrid or plaintext) as a regular email attachment. As for the audio file, if the audio file is larger than 20 MB, then give send it as a link, such as link to Google Drive or similar - but make sure it doesn’t require share access - it needs to be a simple link that the RA can just click and immediately get to your audio file - please nothing that requires other steps like logging in, or share required, or making an account, etc. For example, you could save it to Google Drive and select "Anyone with link can view" and give that link to the RA.
 
In the body of the email to the RA, tell us what you want for these options:

Filter out stop-words?  Yes (Default) OR  No

Filter out unstressed vowels?  Yes (Default) OR  No

Filter out vowels whose F1 or F2 bandwidths are over 300?  Yes (Default) OR  No

Voice type: low OR high

Your e-mail address:     
We will not store or distribute your address.

Which DARLA function should we run for your job?
Choose one:

-Run my job in the regular TextGrid method (recommended) Here is the how-to video

OR

-Run my job in the Plaintext method (not recommended but we still have this feature. It's just that the alignment sometimes fails due to noises or pauses in a long plaintext transcription, whereas you can avoid those noises if using TextGrid method since you put boundaries only around each chunk that needs to be processed, not the noises)

OR

-Just do the vowel extraction step (this is for people who have already made manual corrections to a previously aligned TextGrid from DARLA output, and now want to resubmit their corrected version just for vowel extraction (along with the original audio file)


(4) If you want to use DARLA's Deepgram/Bedword (the completely automated function), please send the RA your Deepgram key number, along with you audio file. RA's email: beepboopbeepboop28@gmail.com



===============================================

That's all for now. Very sorry that DARLA isn't directly available to the public, but this is the best we can do under the circumstances. Please let me know if you have any questions: James.N.Stanford@Dartmouth.edu
Jim

------------------------
James N. Stanford
Linguistics Department
Dartmouth College
Hanover, New Hampshire USA