Google Cloud Text-to-Speech

Turn text and SSML into natural sounding speech with Google Cloud Text-to-Speech. Agents synthesize audio saved to FlowRunner file storage as a ready URL, render long form content as asynchronous jobs, and browse the voice catalog for localization.

4 actions Custom auth available

Google Cloud Text-to-Speech website ↗ Platform Documentation ↗ Capability data verified 2026-07-25

The weekly release notes are approved in the docs tool

List Voices resolves the current Neural2 or Studio voice for each target language

Synthesize Speech renders a short sample with SSML pronunciation fixes for product names

The editor listens to the sample and signs off on voice and pronunciation before the full render

Synthesize Speech renders the full episode to FlowRunner file storage and returns its URL

The audio link posts to the team channel and drops into the internal podcast feed

The flow verifies the audio file exists and logs the render settings for the next episode

What This Integration Enables

Text-to-Speech is the delivery half of a voice pipeline: everything your workflows write, someone could be listening to instead. Google Cloud Text-to-Speech renders text and SSML into natural speech across Google's full voice catalog, from Standard through WaveNet, Neural2, Studio, Journey, and Chirp, and this connector removes the plumbing question: Synthesize Speech saves the audio to FlowRunner file storage and hands back a URL any downstream step can use. It also pairs naturally with [Google Cloud Speech-to-Text](/integrations/google-cloud-speech) on the listening side: one flow transcribes what customers say, the next one answers out loud, and both run under the same workflow governance. - Voice meeting digests, alerts, and AI-drafted answers and deliver them in chat or email - Generate IVR and phone-system prompts as mu-law or A-law audio straight from a script table - Produce narration for videos, e-learning modules, and demos in dozens of languages - Turn articles and newsletters into podcast-style audio for accessibility - Render chapter-length and book-length text as background jobs into your own Cloud Storage bucket - Control speaking rate, pitch, volume, sample rate, and device-specific audio profiles per render

Without FlowRunner

Written updates go unread The digest that took an hour to write dies in an inbox because nobody reads at a desk anymore

Recording is a bottleneck Narration waits for a person, a microphone, and a quiet room, so audio versions rarely happen

IVR prompts are a project Changing one phone menu line means booking voice talent or living with the robotic default

With FlowRunner

Every written artifact can ship as audio Digests, release notes, and answers render to MP3 with a ready URL the moment they are approved

One script, many languages A localization loop walks List Voices and renders the same script across dozens of languages

Phone prompts render from a spreadsheet IVR scripts synthesize straight to telephony formats, reviewed once, deployed everywhere

Use Case Scenarios

The meeting summary the team actually consumes

When a transcript lands from [Fireflies](/integrations/fireflies-service), the flow summarizes it, passes the summary to Synthesize Speech with a Neural2 voice, and posts the returned audio URL to [Slack](/integrations/slack) beside the text version. People driving between showings or walking a warehouse floor listen instead of scroll. The same pattern voices daily metrics digests and incident postmortems, with the audio file already hosted, no bucket setup required.

IVR prompts from a spreadsheet, not a studio

The phone team keeps menu scripts in [Google Sheets](/integrations/google-sheets). When a row changes, the flow renders the script with Synthesize Speech using a mu-law telephony profile at the sample rate the phone system expects, files the audio, and posts the link for review. Fifty prompts across three languages render in one pass, each voice picked from List Voices by language tag, and the review happens once against a written script instead of fifty times against a microphone.

The audiobook-length render that runs itself

A finished training manual needs a narrated edition. The flow calls Synthesize Long Audio with the full text and a gs:// output path in the team's own bucket via [Google Cloud Storage](/integrations/google-cloud-storage), then polls Get Long Audio Operation until done is true. The finished WAV gets a signed URL and lands in [Notion](/integrations/notion) next to the written module. Nothing blocked while Google rendered, and the flow handles a failed job by surfacing the error field instead of silently shipping nothing.

Human-in-Loop Highlight

The voice you synthesize is the voice your customers hear, and a full library render is a spend you make all at once. When the team decides to move the e-learning catalog from a WaveNet voice to a Studio voice, the naive automation re-renders every module immediately: hours of premium-tier synthesis billed per character, producing audio nobody has heard. The FlowRunner pattern splits the decision from the batch. The agent renders one representative module with Synthesize Speech, posts the sample, and waits. A person listens for the mispronounced product names and the pacing that reads fine on paper but drags out loud, fixes them with SSML, and approves. Only then does Synthesize Long Audio walk the catalog, rendering every module into the bucket at machine speed. The agent does the hundred renders; a human ear signs off on the one that defines them.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via preferred channel

Human decides

Agent resumes with decision

Agent Capabilities

4 actions

Synthesis

Synthesize Speech Converts text or SSML into natural speech and saves the audio to FlowRunner file storage, returning its URL. Choose any voice in Google's catalog or let Google pick by language and gender, with output in MP3, WAV, OGG Opus, mu-law, A-law, raw PCM, or M4A, plus speaking rate, pitch, volume gain, sample rate, and device audio profiles. The synchronous endpoint accepts at most 5,000 bytes per request, with SSML tags counting toward the limit.

Voice Catalog

List Voices Lists every voice available for synthesis, optionally filtered by language with partial matching, so en returns all English variants and en-GB only British ones. Each entry reports the voice name, supported BCP-47 language tags, gender, and natural sample rate. The lookup step behind localization loops and voice selection.

Long Form

Synthesize Long Audio Starts an asynchronous synthesis job for inputs up to roughly 1 MB of text or SSML, far beyond the synchronous limit. Google writes LINEAR16 WAV output directly to a Cloud Storage bucket you own, so the gs:// output URI and bucket IAM roles must be in place first. Returns a long-running operation immediately.
Get Long Audio Operation Retrieves the state of a long-audio job: progress percentage while running, the finished file at your Cloud Storage URI once done is true, and a described error on failure. Poll periodically rather than continuously.

Frequently Asked Questions

What can FlowRunner do with Google Cloud Text-to-Speech?

FlowRunner agents can run Synthesize Speech, List Voices, and Synthesize Long Audio in Google Cloud Text-to-Speech, plus 1 more action.

Does connecting Google Cloud Text-to-Speech to FlowRunner require OAuth?

Google Cloud Text-to-Speech uses a custom authentication method to connect to FlowRunner.

Can Google Cloud Text-to-Speech trigger a FlowRunner workflow automatically?

Google Cloud Text-to-Speech doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.

Start building with Google Cloud Text-to-Speech

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us