Google Gemini: Gemini 3.8 Flash-Lite TTS
Google Gemini 3.8 Flash-Lite TTS (GA): fast, cost-efficient text-to-speech for high-throughput production, real-time voice-agent cascades and read-aloud features, sharing the 3.8 Flash TTS schema, 30 prebuilt voices, Extended Voice Library, voice design and voice replication.
ストリーミング: PCM · µ-law · A-law
プロバイダー
同じモデルを複数のプロバイダーがホストしている場合があります。バックエンドを固定したい場合は1つのプロバイダーを選び、そうでなければルーティングに選択を任せられます。
対応言語
ace-Arab
af
ak
am
hy
as
awa
ban
bn
bjn-Latn
eu
be
bho
bs
bug
bg
yue
ca
ceb
ckb
hne
zh-Hans
zh-Hant
hr
cs
da
nl
arz
en
et
fil
fr
gl
lg
ka
de
el
gu
ht
khk
ha
he
hi
hu
is
ilo
id
pes
it
ja
jv
kam
kn
ks-Arab
ks-Deva
kk
km
ki
rw
kg
ko
ky
lo
ln
mk
mag
mai
ml
mt
mni
mr
min-Arab
lus
npi
fuv
azj
nso
uzn
nb
nn
ny
ory
prs
pl
pt
pa
ro
ru
sat
sr
si
sk
azb
pbt
es
arb-Arab
arb-Latn
lvs
zsm
ta
te
ボイス
機能
Generates spoken audio from a verbatim text transcript via the Interactions API (POST /v1beta/interactions, response_format.type 'audio') or generateContent (speech_metadata per part).
text_to_speech
One-shot synthesis returns the whole clip (WAV by default) in a single response.
synchronous
stream: true on the Interactions API returns SSE step.delta audio chunks (headerless 24 kHz PCM by default) as they are synthesized.
streaming
Supports 101 documented languages; the model auto-detects the input language (voices are language-agnostic).
multilingual
Multi-speaker dialogue with up to 2 prebuilt-voice speakers per request (speech_config.speakers + per-turn speech_metadata.speaker; mode 'conversational').
multi_speaker
Turn-level delivery (emotion, pace, volume, accent) via speech_metadata.style, point-in-time vocal events via inline angle-bracket tags, and inline IPA pronunciation overrides.
style_control
Custom vocal persona generated from a natural-language description (POST /v1beta/voices type='prompted'), returning a persistent voice_... id; 200 stored voices per project, 1-year TTL.
voice_design
Replicates a speaker from reference plus consent audio (POST /v1beta/voices type='replicated') as a persistent voice_... id or a stateless 7-day voicekey_... .
voice_cloning
コード
ElevenLabs SDKでのストリーミングTTS
ElevenLabs SDKをAllModels経由でルーティングし、音声レスポンスをストリーミングします。
import { ElevenLabsClient } from "elevenlabs";
const client = new ElevenLabsClient({
apiKey: process.env.ALLMODELS_API_KEY,
baseUrl: "https://api.allmodels.io/el"
});
const stream = await client.textToSpeech.stream("voice_id", {
text: "I'm sorry, Dave. I'm afraid I can't do that.",
modelId: "gemini/gemini-3-8-flash-lite-tts",
outputFormat: "mp3_44100_128"
}, {
queryParams: {
provider: {
"only": [
"gemini"
]
}
}
});
const reader = stream.getReader();
for (;;) {
const { done, value } = await reader.read();
if (done) break;
// Write value to your response, file, or audio playback pipeline.
}