← Live workflow · Open K.A.T.E.

K.A.T.E.’s voice studio

Original direction: warm American, calm, clear, lightly witty.

Listen to references

These public-domain clips share one reference speaker from LJ Speech. They demonstrate recorded speech, not K.A.T.E.’s final voice. Source and license

Record an original voice

Read naturally in a quiet room, keeping the microphone position consistent. This starter script is not enough to train a finished voice.

Microphone off. Recording starts only when you click.

Each take is limited to 20 seconds. Audio stays in this tab until downloaded; it is not uploaded. Explicit recording metadata (ID, consent, dates, time zone, duration and outcome) is saved to Home Base. A visible indicator is always shown; system start/stop notifications require browser permission. Save takes before closing or refreshing.

Speech model status

No K.A.T.E. speech model has been trained. Original recordings, a local speech-training runtime and listening tests are still needed. The experimental text model cannot generate audio. Changing Zira’s pitch does not create a new neural voice.