Privacy

This site collects speech recordings for a research project: building automatic speech recognition for Burmese. It is used by people who were invited to take part, who signed a consent form on paper, and who sign in with an access code rather than a name. That form is the agreement; this page describes what the software actually does with what it collects.

You were given a copy of that form to keep. If you no longer have it, ask the agency below for another. This site does not hold a copy.

The terms of using this website — signing in, your access code, and what to record — are on the terms page.

Who is responsible

The recordings, and the research they are collected for, belong to Sagar Data and Assets Ltd. (UK), the research agency named on the consent form. Questions about your data, and requests to withdraw or delete it, go to [email protected].

What is collected

Your recordings. The audio you record or upload, as captured — the app does not alter, compress or filter it.

Which question each recording answers. Every file is named with the subject, topic and question codes, plus the group and participant number your access code stands for — for example G003_047. That is the only identity attached to it.

Notes about sending. So that failures can be found and fixed, the server records when an upload started, whether each part arrived, how large it was, how long it took, whether it finished, and — when something goes wrong in the browser that the server never sees — a one-word category for the failure and the time the phone reported it.

A device label, not an address. To tell "the same phone had trouble five times" from "five different phones", each request's network address is put through a one-way cryptographic function with a secret key, and only twelve characters of the result are kept. The address itself is never written down, and the label cannot be turned back into one.

One cookie. A single session cookie, so the site remembers you have entered your code. It expires after eight hours of inactivity, is not readable by scripts, and is sent only over HTTPS. There are no advertising, analytics or tracking cookies on this site — none at all.

Who else sees the request

This site sits behind Cloudflare, a company that carries requests to us and blocks attacks before they arrive. Every visit reaches a Cloudflare machine first, so Cloudflare sees your network address and the ordinary details of the request — which page, which browser, at what time — before this application sees anything at all. What reaches the application is the address as Cloudflare reports it, and that is the address turned into a label as described above. It is not stored either way.

Cloudflare keeps its own records under its own policy, and Amazon Web Services, where this application runs and the recordings are stored, is in the same position. Those are their policies to describe, not ours, and this page does not speak for them.

What is never collected

Your name, your phone number, your address, your email address, your date of birth. The software has no field for any of them, and no page asks for them. The consent form is the only place your name exists, and it is on paper.

Text you type is not collected either: the only things this site accepts from you are your access code and your recordings.

Where it is kept, and for how long

Recordings are stored in Amazon Web Services (Amazon S3), in the United States. The application runs on AWS in the same region. They are kept for as long as the research needs them — the recordings are the dataset, so there is no date after which they are discarded.

The notes about sending, and the device labels, are kept for 7 days and then deleted automatically.

Who else sees it

The recordings may be shared with other researchers under appropriate ethical and data protection guidelines, and used for academic research, for developing and evaluating speech recognition and AI systems, for teaching, and in publications or presentations — in which your identity is not revealed. That is what the consent form says, and it is what happens.

As the consent form also says: if the dataset or the technology built from it is later used commercially, participants may receive further compensation. We are not going to tell you your recordings will never be used for commercial purposes, because the form you signed says they may be.

Nobody sells, rents or shares the notes about sending, or the device labels, with anyone. They exist to fix faults and to stop abuse of the site, and then they are deleted.

Your choices

Taking part is voluntary. You may stop at any time, and you may ask for your recordings to be deleted where that is technically possible. Ask at [email protected] — quote your participant number (the code shown on your screen, like G003_047), never your access code.

Trouble with the site

If the site itself is not working — the page will not load, your access code is refused, or a recording will not send — write to [email protected]. That is about the software, not about your data: questions about your recordings, and requests to delete them, go to the agency above. Never send your access code.

Keeping it safe

The site is served over HTTPS. Access codes are held in AWS Secrets Manager, not in the software. Recordings go to storage that only this application can write to. Signing in is rate-limited, so a code cannot be guessed by trying many.

Changes to this page

If what the software does changes, this page changes with it. It describes the software as it is today, not as it is intended to be.