AI Speak

ai.speak AI v0.1.0

Turn text into spoken audio (TTS) via any OpenAI-compatible /audio/speech endpoint. The audio is emitted as first-class binary (blob ref or inline base64) — pipe it to write-file, an HTTP response, an email, or a voice bot.

The AI Speak step on the Studio canvas — The **AI Speak** step as it appears on the Studio canvas — input pins on the left, output ports on the right.

Finding it in the library

Search the builder's node library for AI Speak (it lives under AI). A single click opens the in-editor docs panel shown here — description, ports, and every property, without leaving the canvas. Double-click (or drag) to add it to the workflow.

AI Speak in the node library, with the in-editor docs panel open — The library entry and the in-editor docs panel for **AI Speak** — the same reference this page is generated from.

Wired up in the builder

AI Speak in a real, runnable flow — captured live from the Studio editor, exactly as it looks on your canvas. The ◈ ports carry the AI Model provider you wire in. This is the same workflow used for the example input & output below.

AI Speak wired into a runnable workflow in the Studio builder — **AI Speak** wired into a runnable flow — input on the left, output on the right, AI Model provider wired into the ◈ ports below.

How it’s configured

The node’s Configure panel as it opens in the builder when you select the step — every setting laid out with real values. Click any field to edit it.

The AI Speak node's Configure panel in the Studio builder — The **Configure** panel for **AI Speak**, showing the settings from the flow above.

Ports

Ports are the node’s contract with its neighbours. In the editor a port label renders bold when wired and italic when optional; ◈ ports accept attachment carriers rather than data wires.

Direction	Port	Label	What flows through it
Input	`input`	Input
Output	`output`	Audio

How data flows through it

AI Speak consumes the content of the incoming envelope — when it is fed directly by a trigger, the trigger’s wrapper is unwrapped at the node boundary so the node sees the actual data, not the metadata shell. Its output becomes the payload for the next node, while the envelope (trace ids, correlation, binary refs) rides along untouched. In the Runs view you always see the whole envelope for both sides of this node.

Expressions in the config

String-typed properties accept {{ }} expressions evaluated against the incoming item at run time — e.g. {{ $json.customer.email }}. On this node that’s voice. JSON- and code-typed fields never interpolate — they are passed through literally.

Build it with AI

Every node in this reference is reachable through Flowdrome’s AI Copilot and the MCP tools — say what you want, and the graph surgery happens server-side. Node types resolve fuzzily, so the catalog label (AI Speak) works as well as the exact type id (ai.speak).

In the Copilot panel (or any connected AI):

add a ai speak node after the trigger

As a step in a create_chain_workflow call:

{"type":"AI Speak","config":{}}

Raw MCP call — add this node to a workflow with add_node

curl -s -X POST http://localhost:48170/mcp -H "content-type: application/json" -d '{ "jsonrpc": "2.0", "id": "1", "method": "tools/call", "params": { "name": "add_node", "arguments": { "workflowId": "<id>", "type": "AI Speak" } } }'

Example input & output

Captured from a real test run of the workflow above — this is what you see in the run data panel after pressing Test workflow.

Input — what the node received

The AI Speak node's input envelope in the run data viewer — The input envelope in the Runs view — Flowdrome always shows the whole envelope, with the payload inside `body`.

Output — what the node produced

The AI Speak node's output envelope in the run data viewer — The output envelope after the step ran.

Property reference

Every setting, with its type and default — the same fields shown configured in the panel above.

Property	Type	Default	Description
Voice `voice`	`string`	`"alloy"`	The speaking voice — whatever the endpoint offers. OpenAI: alloy/echo/fable/onyx/nova/shimmer; Kokoro: af_heart, af_bella, am_adam, bf_emma, … (mixable like "af_heart+af_bella").
Text field `textField`	`field`	`""`	Dot-path to the text to speak. Blank = the whole payload (stringified).
Format `format`	`select`	`"mp3"`	Audio container for the generated speech. `mp3opuswav`
Timeout (ms) `timeoutMs`	`int`	`120000`	Abort the speech call after this many milliseconds.

Using it

Text in, spoken audio out

AI Speak posts to any OpenAI-compatible /audio/speech endpoint and emits the clip as first-class binary: audio (inline base64 for short clips, a blob reference for longer ones), contentType, and the passthrough of your input. Set Text field to the dot-path that holds what to say (reply reads the answer straight off an AI Prompt node) and pick a Format (mp3, wav, opus, …).

Like every model-consuming node, the engine to speak WITH is not typed into the node — wire a ◈ Model carrier into the model port. For a local, no-cloud setup use the generic-url provider pointed at a Kokoro server:

docker run -d -p 8880:8880 ghcr.io/remsky/kokoro-fastapi-cpu:latest

then on the ◈ carrier: provider generic-url, base URL http://localhost:8880/v1, model kokoro. OpenAI’s hosted voices work the same way — base URL https://api.openai.com/v1, model tts-1 or gpt-4o-mini-tts, plus a stored credential.

Voices — a free string, and the Kokoro catalog

Voice is deliberately a free text field, because every backend ships its own list — the node never second-guesses it. OpenAI accepts alloy, echo, fable, onyx, nova, shimmer.

Kokoro voice ids encode language + gender + name: the first letter is the accent/language (a American English, b British English, e Spanish, f French, h Hindi, i Italian, j Japanese, p Brazilian Portuguese, z Mandarin), the second is f/m for female/male. The English sets:

Set	Voices
American female (`af_`)	`af_heart` ★, `af_alloy`, `af_aoede`, `af_bella`, `af_jessica`, `af_kore`, `af_nicole`, `af_nova`, `af_river`, `af_sarah`, `af_sky`
American male (`am_`)	`am_adam`, `am_echo`, `am_eric`, `am_fenrir`, `am_liam`, `am_michael`, `am_onyx`, `am_puck`, `am_santa`
British female (`bf_`)	`bf_alice`, `bf_emma`, `bf_isabella`, `bf_lily`
British male (`bm_`)	`bm_daniel`, `bm_fable`, `bm_george`, `bm_lewis`

af_heart (★) is the flagship — the most natural of the set and the default in the shipped demos. A running Kokoro server lists everything it can do at GET /v1/audio/voices, including the non-English sets. Kokoro also blends voices with + (af_heart+af_bella).

Hearing the result

In the editor: chain a Play Audio node after Speak — the clip gets an inline player in the Console and is saved with the run.
On a page: end a webhook/form workflow with an HTTP Response node in payload mode — the base64-marked audio field is served as raw bytes, so <audio controls src="https://host/webhook/tts?text=hello"> just works.
In a chat: a web-chat workflow whose final payload carries audio speaks its replies — the widget plays them automatically.

Tips

The AI Demo folder that ships with the install has the full pair: 12 — Speak writes ./ai-demo-speech.mp3 and 13 — Transcribe reads it back with local Whisper.
CPU TTS dominates chat latency (~2–3s for a sentence, vs ~0.5s for the LLM). The web-chat serve layer already streams the TEXT reply the moment the model finishes and follows with the voice when it lands — no configuration needed.
Long texts produce clips over 64KB, which travel as blob references between nodes; Play Audio and HTTP Response inline them back to portable base64 automatically.

Related nodes

The rest of the AI group — the same folder you’d scan in the editor’s library.

This page is generated from the node registry by gen-node-docs.mjs on every site build — ports, properties, defaults and visibility rules cannot drift from the code. The screenshots and example data are captured from a live Flowdrome by npm run shots:nodes and npm run gen:examples. The hand-authored notes above are merged in.