No voice is installed on this build. Record or upload a short clip of clean speech and all eight moods will use it.

The engine is still loading in the background. Installing a voice works now; the first take plays once both are ready.

3 to 40 seconds of clean, single-speaker audio — 5 seconds or more with some expression works best. The transcript must match the clip word for word; a mismatch makes every mood worse.

Mood tuning was calibrated on the preset voice, so expressiveness varies with the clip. A seed reproduces a take only under the same voice.