All learning paths Learn See Try Make What a language collection needs What this lesson is about Plan what you are collecting, from whom, and how much, before the first recording is made.
A district team wants Bhojpuri for a voice assistant, and two phones are free today. What goes on paper first?
A The number of hours you want, since a bigger set is worth more later. B Decide the material, the speakers, the amount and the purpose before you record. C A written promise to the district office that the files stay with them.
Check my recall Why this matters A collection with no plan cannot be checked later, because nobody agreed what it was for. The IndicVoices team states its own plan out loud. It reports 7,348 hours of speech from 16,237 speakers, across 145 districts and 22 languages.
Write a one-page plan for a language collection before the first recording is made.
A collection with no plan cannot be checked later, because nobody agreed what it was for. The IndicVoices team states its own plan out loud. It reports 7,348 hours of speech from 16,237 speakers, across 145 districts and 22 languages.
Decide the material, the speakers, the amount and the purpose before you record. A plan makes a collection, not its size. Ten planned hours beat a hundred loose ones. Write the plan where the people recording it can read it, in their language. Task A district team wants to collect Bhojpuri speech for a voice assistant.
Weak approach Start recording whoever is free today, and decide later what the recordings are for.
Stronger approach One page first. Forty speakers across three blocks, twenty minutes each. Free talk about farming and health, for a Bhojpuri voice assistant. The district team holds it.
Why the stronger approach works The strong version names the material, the speakers, the amount and the purpose. That is the first rule, and it is what makes the hours checkable.
Try a changed situation You are no longer collecting speech. A partner offers two photo sets: a big one with no plan, and a small planned one. Which one do you take?
Make something yourself Write a one-page plan for a language collection you could really run. Name the material, the speakers, the amount, the purpose, and who will hold it.
Lesson 1 of 12 on Bharat-language data craft next Take consent in their own language Part 1 ends in a work sample you can send
Read the primary or official source: Javed et al., IndicVoices (arXiv)