How Inzwi works

Voice AI for Shona is built one contribution at a time. Here's how you add yours.

1

Read aloud

Read a short Shona sentence aloud, on WhatsApp or here — or a story, a paragraph at a time.

2

Talk freely

Answer an everyday question in your own words, the way you would say it to a friend.

3

Review

Listen to other contributors' clips and say whether each one is clear and matches its text.

4

Transcribe

Write down what was said in a free-talk answer, so it can be used for training.

Step 1

Read aloud

You are shown a short sentence in Shona and read it aloud. The sentence becomes the transcript, so every clip arrives already labelled. On WhatsApp the bot sends one sentence at a time and you reply with a voice note. On this site you can also read a story, one paragraph at a time.

  • Read at a normal pace, in your own accent
  • Somewhere quiet, phone close to your mouth
  • Re-record before sending if you stumble
  • Works on WhatsApp or in the browser

Read aloud:

"Ndiri kufara chaizvo nhasi"

Step 2

Talk freely

You are asked an everyday question and answer in your own words for fifteen to thirty seconds. This is how people really speak, which reading cannot capture.

  • Answer as you would to a friend
  • Mixing in English words is fine — that is real speech
  • Aim for 15 to 35 seconds

Question:

"Pakati pa Winky D na Jah Prayzah, ndeupi muimbi waunofarira?"

Your answer:

Step 3

Review recordings

Listen to clips other contributors have read and decide whether each one is usable. A clip is accepted or rejected only when two reviewers agree, and a rejection always says why.

  • Is the speech clear and loud enough?
  • Does it match the sentence on screen?
  • Pick a reason when you reject, so the speaker can improve
  • If you share the speaker's dialect, you may be asked whether they sound like it

Should say:

"Ndiri kushanda kubasa nhasi"

AcceptReject
Step 4

Transcribe free talk

Free-talk answers have no text to go with them. Listen to one and type exactly what was said, in Shona, so the answer can be used to train speech recognition.

  • Write what you hear, not what you think they meant
  • Keep English words in English
  • Say how confident you are

You type:

"Ini hangu ndinofarira..."

How quality is kept

Nothing goes into the corpus on one person's word.

Two reviewers agree

Every read-aloud clip needs two matching verdicts from other contributors before it is accepted or rejected.

Audio checks

Each upload is checked for silence, clipping and duplicates before a person ever hears it.

Dialect confirmed

Speakers of the same dialect confirm whether a contributor sounds like the dialect they declared.

Ready to Contribute?

Every recording, review and transcript brings us closer to AI that truly understands Shona.