Audio Generation Node

Audio generation produces spoken voice on the canvas through ElevenLabs. It covers two voice modes: text-to-speech, which narrates a prompt with a chosen voice, and speech-to-speech, which re-voices an input recording while preserving its delivery. Both run through the shared /api/generate-audio route and store the result as durable media.

What it does

• Text-to-speech — converts a prompt into narrated audio with a selected voice and model.
• Speech-to-speech — converts an input recording into the same content spoken by a target voice. The Speech to Speech node reads its audio from a connected audio-in input.
• Decodes the returned audio, stores it durably, and emits it on the audio-out port.

Provider and defaults

Audio generation is bound to ElevenLabs. When the node does not specify a voice or model, the route applies these defaults.

Inputs

• Prompt / text — required for text-to-speech. The route accepts either text or prompt.
• Audio input — required for speech-to-speech, supplied as inline base64 (audioBase64) or a public HTTPS URL (audioUrl). The Speech to Speech node resolves this from its connected audio input.
• Voice — an ElevenLabs voice id. The route enforces voice usage access when shared credentials are used.
• Voice settings — optional stability, similarity boost, style, speed, and speaker boost, each clamped to its valid range.
• Text-to-speech extras — language code, seed, surrounding text for continuity, request id stitching, pronunciation dictionaries, and text-normalization options.

Outputs

On success the node emits MP3 audio on the audio-out port and stores it as durable media. Downstream nodes consume it like any other audio asset, and it remains reusable across sessions.

Generate voice audio

1. 1

   Add a voice node

   Add a Text to Speech or Speech to Speech node. The configuration panel opens automatically so you can choose a voice and model.

2. 2

   Provide the source

   For text-to-speech, type a prompt or connect a text source. For speech-to-speech, connect an audio input on audio-in.

3. 3

   Tune voice and settings

   Pick a voice and adjust stability, similarity, style, or speed. Speech-to-speech can also remove background noise from the source recording.

4. 4

   Run and reuse

   Run the node to generate audio. The stored result flows out of audio-out for downstream steps.

Agent and API notes

Both modes post to /api/generate-audio with a mode field. The route is rate-limited, bound to a canvas for credential scoping, and idempotent per provider operation so a retried run does not double-charge. The examples below show a text-to-speech and a speech-to-speech body.

1{2  "mode": "tts",3  "text": "Welcome to Builder Studio.",4  "voice": "EXAVITQu4vr4xnSDxMaL",5  "model": "eleven_multilingual_v2"6}

1{2  "mode": "sts",3  "voice": "EXAVITQu4vr4xnSDxMaL",4  "model": "eleven_multilingual_sts_v2",5  "audioUrl": "https://example.com/source-recording.mp3"6}