All learning paths Learn See Try Make Check a batch before it goes What this lesson is about Pull a fair sample, listen or read all of it, and count the faults instead of trusting a feeling.
Two thousand Kokborok clips are ready to go out on Friday, and you have an afternoon. How do you check them?
A Pull a fair sample: not the first ten, not the ones you remember. B Take thirty at random, hear the first few seconds of each, and pass it. C Play thirty at random, and send back any the recording app already flagged.
Check my recall Why this matters Common Voice does not argue about a clip. Two other people vote it up or down, and two matching votes settle it. A rule written in advance is what keeps a tired reviewer honest.
Check a finished batch with a fair sample, a fault count, and a send-back rule set in advance.
Common Voice does not argue about a clip. Two other people vote it up or down, and two matching votes settle it. A rule written in advance is what keeps a tired reviewer honest.
Pull a fair sample: not the first ten, not the ones you remember. Count the faults by kind. A count is a finding; a feeling is not. Set the send-back rule before you look, not while you are under pressure. Task Two thousand Kokborok clips are ready to hand over on Friday.
Weak approach Listening to the first twenty clips, hearing nothing bad, and sending the batch.
Stronger approach Thirty clips drawn at random, all thirty heard, faults counted: nine cut short, four wrong speaker, two silent. The rule said send back above five in a hundred.
Why the stronger approach works The first twenty clips are the ones recorded most carefully. A random sample is the only one that tells you about the batch.
Try a changed situation The batch is now four hundred photos of shop signs, not audio. Your reviewer says the batch feels untidy. What do you ask for?
Make something yourself Check one real batch. Write how you drew the sample and the fault counts by kind. Then give your send-back decision against the rule you set first.
Lesson 8 of 12 on Bharat-language data craft next Write the sheet that travels with it Part 3 ends in a work sample you can send
Read the primary or official source: Ardila et al., Common Voice (arXiv)