ACADEMY / AI VOICE

Build an AI Voice workflow that can survive retries, QA and release review.

This lesson uses a real FC-EL-011 lineage. You will inspect the script, provider state, bad output, improved output and the acceptance decision.

45–60 minNo prerequisiteDeliverable: QA-reviewed candidate
Start lesson
WHAT YOU WILL BUILD

Script → provider → retry → QA → release state.

01

Version the script

Separate copy changes from provider retries.

02

Generate with a real provider

Inspect ElevenLabs UI and usage state.

03

Compare bad vs good output

Listen for rhythm, clarity and fit.

04

Run QA

Reject defects before release.

05

Record the state

Keep the accepted candidate and evidence linked.

Real ElevenLabs provider UI
PROVIDER LAB

Inspect the real provider state, not a mockup.

Credits and requests are not the business outcome, but they matter when diagnosing retries and cost.

  • Identify the provider request state.
  • Separate retry cost from useful output.
  • Decide whether the next action is regenerate, repair script or stop.
LISTEN & COMPARE

Bad output and improved output should sound different without reading an audit log.

V1 / REJECTED SAMPLE

Focus check: pacing, pronunciation and business-use fit

Write down one defect you would reject before publishing.

V6 / QA-ACCEPTED SAMPLE

Compare the same checks after correction

Compare pacing, intelligibility and business-use fit.

YOUR DELIVERABLE

Finish with a release decision, not a generated file.

Script versionWhich copy was actually rendered?
Provider evidenceWhat request or usage state supports the run?
QA resultWhat defect was rejected or accepted?
Release stateReady, blocked or Human review required?
Run the 6-check QA exercise