Teach a device to wake up when you talk to it

About the project

"Hey, Mira. Take a photo." For a voice assistant to catch those words the first time, every time — across loud rooms, soft voices and fast talkers — it has to hear them thousands of ways first.

This project records exactly that. You'll record eight short wake-word phrases, each 50 times across controlled variations: normal, loud, soft, fast and slow. It's methodical, satisfying, checklist-style work you can finish in a single day, entirely from home.

What you'll do

• Record 8 short phrases ("Hey, Mira", "Take a photo", "Play music", etc.)
• Capture 50 takes per phrase across five speed/volume styles (10 each)
• Deliver 400 clean WAV recordings following the naming convention
• Typical completion time: one focused day

Who we're looking for

• Fluent or native English speakers based in the United States
• Consistent, clear pronunciation
• People who can follow a precise technical spec without drifting

Requirements

• A quiet indoor recording space — no background noise, clicks or echo
• A device that records WAV: 16-bit, 16 kHz, mono
• No noise suppression, enhancement or AI processing on the files
• Each clip must contain only the required phrase
• Paid per accepted submission set

How it works

1. 1.Create your account

   Sign up free with your email or Google. Takes about a minute.

2. 2.Complete your profile

   Tell us your languages, location and setup so we can match you.

3. 3.You're automatically applied

   Sign up from this page and your application to this project is submitted for you.

4. 4.Our team reviews and onboards you

   Applications are reviewed on a rolling basis and we follow up by email.

Frequently asked questions

Do I need professional recording equipment?

No studio needed. Most modern laptops and phones with a decent free recording app can produce 16-bit/16 kHz mono WAV files. What matters most is your room: quiet, soft surfaces, no hum.

Why 50 takes of the same phrase?

Variation is the product. A wake-word model trained only on one careful delivery fails on a sleepy whisper or an excited shout. Your ten soft takes are as valuable as your ten loud ones.

What gets a submission rejected?

Background noise, clipped audio, wrong file specs, missing takes or files that break the naming convention. Follow the checklist we provide and acceptance is straightforward.