Voice Workflow Preparation
Prepare authorized reference audio, consent records and a short test script before starting an AI voice cloning workflow. Verify which voice modes and models are active in your Astorie workspace first, then evaluate pronunciation and tone on a short sample before recording full narration.
Voice workflows depend on both legal permission and what the active audio interface actually supports. Comparing an unverified recording session with a structured preparation checklist helps you avoid recording unusable samples or assuming a model page equals a live cloning upload control.
Start with documented speaker permission and verify the current Audio node capabilities in your workspace at each stage.
Obtain clear permission from the speaker for the exact narration, channels and commercial or internal uses your project requires. Do not upload or imitate anyone's voice without authorization.
When your workflow uses a reference recording, capture steady, natural speech from one speaker in a quiet space without music, reverb or cross-talk. Match the calm or energetic delivery style that the final script will need.
Open the Astorie canvas and inspect the Audio node controls available to your account. Current Audio node documentation covers Text to Speech, Voice Design, Sound Effects, Music and Video to Audio, while provider pages may describe external model features that are not yet wired into the live runtime.
Enter a two- or three-sentence test script inside the Audio node and generate a sample using an available voice option. Listen on headphones for timbre similarity, pacing, breaths and pronunciation before producing longer segments.
Evaluate three boundaries before relying on any generated voice track in a production timeline.
A model overview page can describe a provider's broader capabilities without proving that every mode is active in the canvas today. For example, the Fish Audio S2 model page notes that the model is not yet integrated into the production runtime. Always confirm what appears inside your live Audio node.
Short test phrases often hide how a voice handles product names, acronyms, dates and conversational pauses. Include your hardest terms in the first test run and rewrite awkward clauses into shorter, speakable sentences.
Keep a clear record of who authorized the voice reference and where the generated audio may be published. Review platform disclosure rules and commercial terms before using synthetic speech in client, broadcast or public work.
Verify permissions, active node inputs and review criteria before producing narration tracks.
| Input | Authorized speaker consent, a short test script entered in the Audio node and any reference input supported by your active workspace. | |
|---|---|---|
| Output | Generated speech audio for listening review; available voice creation or selection modes depend on current Audio node controls. | |
| Review | Check live feature availability in the Audio node, speaker likeness, pronunciation of proper nouns, pacing and audio cleanliness. | |
| Plans and credits | Audio generation uses workspace credits according to the active model and plan tier; check the pricing page for current access rules. | |
| Rights and consent | Use only voices you have explicit permission to represent, and verify commercial-use and disclosure requirements before publishing. |
Review documented Audio nodes and live speech models on Astorie when planning voice tracks.
Check the documented inputs and modes for Text to Speech, Voice Design, Sound Effects, Music and Video to Audio.
Explore supported ElevenLabs speech, narration and voice-design models available for node workflows.
Confirm plan tiers and credit rules before running multi-take audio tests.
Public Audio node documentation confirms Text to Speech and Voice Design from text descriptions, while some provider pages such as Fish Audio S2 state that runtime integration is not yet active. Check the Audio node in your own account to confirm whether reference-audio cloning is available before promising a cloned voice deliverable.
Voice Design creates a custom synthetic voice from a written description of vocal traits inside the Audio node. Voice cloning aims to match a specific real speaker from an authorized reference recording and requires strict consent and a supported reference-audio workflow.
Use a clear recording of one speaker in a quiet room with consistent microphone distance and natural pacing. Avoid background music, heavy compression, room echo or clips where multiple people speak over one another.
Commercial use requires both an eligible paid plan and full legal permission for the voice and script you use. Never clone or imitate a person's voice without their explicit authorization, and check current platform and provider terms before release.
Verify the voice options available in your Audio node, test a short approved script and review the audio carefully before full production.
