Gemini 3.8 Flash-Lite TTS

Gemini 3.8 Flash-Lite TTS — AI Text-to-Speech (Audio Generation)

What is Gemini 3.8 Flash-Lite TTS?

Gemini 3.8 Flash-Lite TTS is Google's high-throughput text-to-speech model in the Gemini Audio family. It turns written scripts into expressive, natural-sounding speech at scale, and instead of reading a script flatly it performs it: you direct delivery in plain English and shift tone, pacing, and emotion line by line. It offers 30 named voices, native two-speaker dialogue, inline vocal cues, and more than 100 languages. On Hume AI's Overall Quality Index it ranks second, behind only the flagship Gemini 3.8 Flash TTS, with major long-form stability gains over Gemini 3.1 Flash TTS.

Key Features

• •Directable expressiveness — a plain-English style field and a temperature control shift register, pacing, and emotion across the whole read.
• •Inline vocal cues — drop <laughs>, <sigh>, <gasp>, and <short pause> into the text and they render where you place them.
• •Native two-speaker dialogue — set Voice 1 and Voice 2 and prefix each line to stage a two-person conversation from one script.
• •30 voices, 100+ languages — strong blind-preference results in Brazilian Portuguese, Japanese, Hindi, Mexican Spanish, Vietnamese, and Modern Standard Arabic.
• •Long-form stability — consistent timbre and natural pacing across extended passages with minimal speaker drift.

Best Use Cases

Gemini 3.8 Flash-Lite TTS fits high-volume dubbing and media localization, audiobook and podcast narration, read-aloud features, and expressive real-time voice agents. Two-speaker staging makes interviews, explainers, and dramatic scenes simple to produce from a single script.

Prompt Tips and Output Quality

Write natural sentences and use the style field for direction such as "cheerful and friendly" or "whispered urgently". Keep numbers and proper nouns in context; the model pronounces them cleanly. Raise temperature to 0.8–1.4 for emotional delivery and lower it to 0.2–0.5 for steady narration. For other languages, write the script natively in that language. Output is a 24 kHz mono WAV, watermarked with SynthID.

How to Use Gemini 3.8 Flash-Lite TTS

Gemini 3.8 Flash-Lite TTS converts a text script into expressive speech. The only required field is text; everything else shapes who speaks and how. The fastest route to a great read is to pair a fitting voice with a plain-English style direction and a temperature that matches the energy you want. Inline cues — <laughs>, <sigh>, <gasp>, <short pause> — render wherever you place them in the text.

Recommended Settings

• •Narration / read-aloud: one voice, style like "clear, friendly narration", temperature 0.3–0.5 for a steady, consistent read.
• •Dramatic or emotional delivery: temperature 1.0–1.2 with a style that names the arc ("build tension, then an excited climax"); energetic voices such as Puck or Fenrir suit high-energy reads.
• •Two-speaker dialogue: set Voice 1 and Voice 2, give each turn its own line prefixed with the voice name, and keep temperature near 0.8 for natural turn-taking.
• •Whispered or calm scenes: a softer voice such as Enceladus, style "whispered, barely above a breath", temperature ~0.8 so delivery stays controlled.
• •Other languages: write the script in the target language; no language flag is needed.

Example voices and prompts to try

Each example below ran end to end at the settings shown; the script, voice, style, and temperature carried straight through to the 24 kHz WAV output.

• •Expressive single-speaker (default) — voice Puck, temperature 1.1. Prompt: "Here we go — the final hundred of the women's 400. <short pause> Delgado's in lane four, half a stride back... can she find another gear? And she does! <gasp> She's flying now — 3, 2, meters to go — Delgado takes it! A new national record — 49.06! The Halden Arena is on its feet." The read opened measured and tense, then burst into a shouted finish with the cues and numbers landing cleanly. Tip: put the whole emotion arc into the style field.
• •Two-speaker dialogue — Voice 1 Kore, Voice 2 Charon, temperature 0.8. Prompt: "Kore: So the probe finally reached the vent field — how deep are we? / Charon: Two thousand, four hundred meters. And here's the part nobody saw coming. / Kore: Don't keep me waiting! / Charon: The water is 360 degrees, and something down there is thriving. <laughs> A whole colony of tube worms." Two clearly distinct voices traded turns with natural pacing and a mid-line laugh. Tip: prefix every line with the exact voice name.
• •Multilingual (Brazilian Portuguese) — voice Aoede, temperature 1.0. Prompt: "Hoje a gente faz um brigadeiro de verdade! <short pause> Duas latas de leite condensado, quatro colheres de chocolate em pó, e muito carinho. Mexe sem parar até desgrudar do fundo da panela. Sentiu esse cheiro? <laughs> Perfeito — agora é só enrolar e cobrir com granulado." A warm, upbeat recipe read with natural pronunciation. Tip: describe the persona in style and write the script natively in the language.
• •Style range, whispered suspense — voice Enceladus, temperature 0.8. Prompt: "Keep your voice down. <short pause> We are standing in the oldest room of the abbey — 800 years, and not one stone has moved. They say the last monk never left. <sigh> Do you hear that... just beyond the door? Don't turn around yet." A hushed, breathy delivery that stayed quiet and controlled throughout. Tip: set the register in style and keep temperature moderate so the voice doesn't drift loud.

Further reads the model is built for: "a calm customer-service agent confirming an order number", "an audiobook narrator with a distinct regional cadence", "a two-host podcast cold open with overlapping energy".

Does it support multiple speakers?

Yes. Set Voice 1 and Voice 2 and prefix each line with the voice name to stage a two-person conversation from one script.

How many voices and languages are available?

Thirty named voices across more than 100 languages, with especially strong quality in Brazilian Portuguese, Japanese, Hindi, and Mexican Spanish.

How do I add emotion or a delivery style?

Use the plain-English style field and inline cues like <laughs>, and raise temperature for more expressive reads.

Can it clone or design a custom voice?

Not on this endpoint — voice design and replication belong to Gemini 3.8 Flash TTS; Flash-Lite uses the 30 preset voices.

How is it different from Gemini 3.8 Flash TTS?

Flash-Lite is the high-throughput production tier with the same line-by-line direction controls; Flash adds voice design and replication.