Auditoría de pronunciación

---
name: pronunciation-audition
description: "Identify pronunciation risks in a script and generate speech samples for reviewed wording variants."
---

# Pronunciation Audition

Identify pronunciation risks in a script and generate speech samples for reviewed wording variants.

## Required Aident Actions

- <action-tag>api:elevenlabs:text_to_speech</action-tag> (required inputs: actionName, voice_id, text): Use this integration action for text to speech operations in ElevenLabs API Documentation.

Use Aident Loadout to read the current Action schema before constructing inputs. Check the required integration connection in Aident Vault. If an Action is billable or mutating, run Aident preflight, show the affected target and quoted cost or risk, and wait for explicit user confirmation before execution.

## Workflow

1. Confirm the requested outcome, source material, destination, audience, constraints, and acceptance criteria.
2. Apply the source-derived guidance below to create a concrete plan. Resolve ambiguity before invoking an Action.
3. Select only the Aident Actions whose documented effect directly advances the requested outcome. Do not invoke every listed Action by default.
4. Inspect the current schema and prepare the minimum valid input for each selected Action.
5. Preflight each selected Action. Execute it only after any required confirmation, in dependency order, and carry returned IDs or asset URLs into later steps.
6. Verify the returned IDs, URLs, statuses, or artifacts against the acceptance criteria. Report partial completion precisely and do not repeat paid or mutating calls blindly.

## Source-Derived Guidance

- Collect the script, target language and accent, selected voice, names and specialist terms, and any known pronunciation preferences.
- Flag ambiguous terms and prepare minimal wording or spelling variants without changing the intended meaning.
- Use api:elevenlabs:text_to_speech to generate comparable samples of the approved variants with the same voice settings.
- Review intelligibility, stress, rhythm, and consistency, then select or revise the strongest rendering.
- Return the selected text, pronunciation notes, and generated sample without claiming authoritative linguistic certification.

## Execution Boundaries

- Treat upstream provider-specific commands as background knowledge only. Execute the workflow through the exact Aident Actions above.
- Never request raw credentials in chat. Use Aident Vault connection flows for required accounts.
- Preserve user-provided wording, brand constraints, rights restrictions, and target identifiers. Do not invent authorization.
- Pass assets by Aident asset ID or supported URL fields. Do not expose caller-local filesystem paths to hosted Actions.
- Stop when a required integration is disconnected, preflight rejects the input, the target is ambiguous, or the user declines a required confirmation.

## Output

Return the execution plan, selected Action names, confirmed targets, Aident result identifiers or asset URLs, verification evidence, and any remaining blocked step.

## Attribution

This Skill adapts the reviewed upstream workflow. See [UPSTREAM.md](UPSTREAM.md) for the pinned source and [LICENSE.txt](LICENSE.txt) for the preserved license.