Google Gemini

The Gemini speech surface is sync-only and rides on the same :generateContent endpoint the language-model package uses. Useful when you already have a Gemini key and don’t need streaming.

For language-model use of Gemini (chat, tools, thinking budget) see Providers / Gemini. For music generation with Lyria see Music generation / Lyria. This page covers only speech.

Install

pnpm add @effect-uai/core @effect-uai/google effect

Layers

Layer	Registers	Capability markers
`@effect-uai/google/GeminiTranscriber`	`GeminiTranscriber` + `Transcriber`	— (sync only)
`@effect-uai/google/GeminiSynthesizer`	`GeminiSynthesizer` + `SpeechSynthesizer`	— (sync only; no `streamSynthesis` either)

import { Config, Effect, Layer } from "effect"
import { FetchHttpClient } from "effect/unstable/http"
import { layer as transcriberLayer } from "@effect-uai/google/GeminiTranscriber"
import { layer as synthLayer } from "@effect-uai/google/GeminiSynthesizer"

const gemini = Layer.unwrap(
  Effect.gen(function* () {
    const apiKey = yield* Config.redacted("GOOGLE_API_KEY")
    return Layer.mergeAll(transcriberLayer({ apiKey }), synthLayer({ apiKey }))
  }),
)

const mainLayer = gemini.pipe(Layer.provide(FetchHttpClient.layer))

Neither layer ships SttStreaming or TtsIncrementalText. Calling streamTranscriptionFrom or streamSynthesisFrom against this Layer is a compile-time error — for streaming, use OpenAI Realtime, ElevenLabs, or Inworld.

Models

STT (audio understanding)

Gemini’s “transcription” isn’t a dedicated endpoint — it’s the multimodal model receiving an audio part plus the prompt “Transcribe verbatim.” Any model that accepts the audio modality works: gemini-3-flash-preview, gemini-3.1-pro-preview, gemini-3.1-flash-lite-preview, gemini-2.5-flash, gemini-2.5-pro.

Limits:

No word timestamps — wordTimestamps: true fails with AiError.Unsupported. The model can emit MM:SS timestamps inside the text if you ask in the prompt, but they’re unstructured.
No diarization — diarization: true fails with Unsupported.
No language forcing — autodetected; language hint accepted but not enforced.
20 MB inline-data ceiling — use the Files API for longer audio (not yet wired in this adapter).

Audio formats accepted: audio/wav, audio/mp3, audio/aiff, audio/aac, audio/ogg, audio/flac.

TTS

Model	Notes
`gemini-2.5-flash-preview-tts`	Steerable preview model
`gemini-2.5-pro-preview-tts`	Higher quality, slower
`gemini-3.1-flash-tts-preview`	Newer flash preview

30 prebuilt voices — Kore, Puck, Zephyr, Enceladus, Charon, Aoede, Leda, Orus, Callirrhoe, etc. (full list in @effect-uai/google models.ts). GeminiVoiceName is a literal-only union with no (string & {}) escape — there’s no cloning path on this surface.

Output is PCM s16le at 24 kHz, returned wrapped in a WAV RIFF header by the adapter (so it drops into an <audio> tag directly).

Request shape

// STT
type GeminiTranscribeRequest = {
  readonly model: GeminiSttModel
  readonly audio: AudioSource
  readonly language?: string // hint only — Gemini autodetects
  readonly prompt?: string // appended to the "transcribe verbatim" instruction
}

// TTS
type GeminiSynthesizeRequest = {
  readonly model: GeminiTtsModel
  readonly voiceId: GeminiVoiceName // literal-only — no cloning
  readonly text: string
  readonly outputFormat?: AudioFormat // ignored — always wav/pcm 24 kHz
  readonly styleInstructions?: string // prompt-level prosody direction
}

styleInstructions is free-form natural language — "say this in a warm, conversational tone", "emphasize the second sentence". Combine with inline prosody tags in the text ([whispers], [shouting]) for finer control. No SSML, no speed / pitch knobs.

Wire / auth notes

Both endpoints are POST https://generativelanguage.googleapis.com/v1beta/models/{model}:generateContent with generationConfig.responseModalities set to ["AUDIO"] for TTS, or with audio supplied as inlineData for STT. The same apiKey works for everything.

Errors

Standard HTTP → AiError mapping. Additionally:

Request shape	Error
`wordTimestamps: true`	`AiError.Unsupported`
`diarization: true`	`AiError.Unsupported`
Audio larger than 20 MB inline	`AiError.InvalidRequest`