Gemini 3.8 text-to-speech

This is an unofficial, independent skill. Gemini and Google AI Studio are Google products; their models, services, documentation, and marks remain Google's. The original skill text and helper code are MIT licensed by RobTar97.

Use the Gemini API v1beta/interactions TTS format for both gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts. Keep the words to be spoken in text, the delivery across a turn in speech_metadata.style, and speaker identity in speech_config. The models share the same request schema.

Choose the model

• Choose Flash for nuanced acting, challenging accents or pronunciation, long narration, heavy use of vocal bursts, and intricate two-speaker overlap.
• Choose Flash-Lite for high-throughput read-aloud, dubbing, and voice-agent output when its performance is adequate. Test a difficult dialogue on Flash if Lite misses timing or emotion.
• Use these TTS models for a prepared transcript. A genuinely live, bidirectional conversation with incoming audio and barge-in belongs to the Gemini Live API. Pipe backchannels here are scripted overlap, not an event-driven interruption mechanism.

Compose the request