What the work involves
This is a single-session speech data task, not an ongoing engagement. You receive a short script and a step-by-step recording guide, record yourself reading it aloud in English, and upload the file through the platform. Start to finish, contributors report the whole thing taking 15 minutes or less. Pay is flat: $50 for one accepted submission.
The research value here is accent authenticity. The dataset is being built to train or evaluate speech models on French-influenced English, which means the recording has to reflect how you actually speak — not a performance. Contributors who put on an accent are screened out, and submissions that sound imitated are typically rejected at review.
What the platform screens for
- Genuine French-influenced English. Expect to be asked where you grew up, what language you spoke at home, and how long you have been speaking English. A live or recorded speech sample is common.
- Recording conditions. A quiet room, a working microphone, and no background noise, music, or overlapping speakers. Phone microphones are fine; the bar is clean audio, not studio equipment.
- Instruction-following in English. The guide is written in English and there is no hand-holding. You need to read it, follow it exactly, and submit in the requested format.
- Consent and usage. You will be asked to agree that your voice recording can be used as training or evaluation data.
Logistics
Fully remote and fully asynchronous — record whenever you like within the project window. Because it is a quick-turnaround project, the window is short: applicants who take days to respond often find the batch closed. There is no interview beyond the screening, no scheduled calls, and no recurring commitment, though contributors who submit clean audio are frequently invited back for later batches.