Voices · Open speech for Sayso

Lend Sayso your voice.

Read ten short cards aloud, in about five minutes. Your recordings teach the first Sayso model to get names, paths, and terms right, on your own computer.

Start reading

No account.
You choose what we publish.

The ledger · updated every minute

Every clip counts.

When the ledger reaches 20,000 clips from 500 voices, Sayso trains its first model on them. The model and the dataset are open.

0of 20,000 clips
0of 500 voices
0 hof speech, so far
4,000The test set is full
12,000First test model
20,000Sayso Solo 1

NEEDED NOW

    Your next cards come from these. They count double.

    How a session works

    Four kinds of cards.

    A session is ten cards and takes about five minutes. You hear each clip before you keep it, and you can skip any card.

    Real speech is the point. Say “um”, stop, start again. Sayso has to learn how people really talk.

    • 1 · Read10 pts

      Ask Priya Raghunathan to review pull request 2318.

      Read the card as it is written.

    • 2 · Say it your way20 pts

      Tell the team that the Kubernetes upgrade moved to Friday.

      Use your own words, then fix what we heard.

    • 3 · Change your mind20 pts

      Give a time for the standup, then correct it.

      Correct yourself out loud, as you would at work.

    • 4 · Spell it15 pts

      Say the name Siobhán, then spell it.

      Letter by letter, at your own speed.

    The game

    Stamps, not confetti.

    Each day you read, your card gets a date stamp. Each kind of card you finish earns an ink stamp. Points stay pending until we check the clip, so only real speech counts.

    Choose whether you show on the weekly leaderboard. The name there is a random one, like juniper-otter.

    Your data

    What we keep.

    You must be 18 or older. Before your first card, you choose whether your recordings only train Sayso models, or also go into the public dataset.

    Read the full consent text

    Your recordings
    Kept, to train Sayso models
    The text you fix for a card
    Kept, as the correct text
    Accent, age range, microphone
    Only if you add them
    Your name, email, or an account
    Never asked
    The public dataset
    Only if you choose it

    Your contributor key deletes everything you recorded. The models are open, so a model that is already released keeps what it learned.

    Read your first card.

    Start reading

    Works in any current browser. A laptop or desktop microphone is best.