Cost-to-Approved-Audio
Character price alone does not tell a buyer what AI voice really costs. VoicePilot measures what it takes to reach an approved result: generation spend, failed generations, revision attempts, human review time, editing time and workaround burden.
Provider observations
No verified observation is treated as zero. Until equivalent runs clear the same approval gate, the system shows missing evidence instead of inventing a cheap winner.
Why this is more useful than price per character
A provider can be cheap per million characters but expensive in production if names fail, rupee amounts need multiple retries, Hinglish requires spelling hacks, or an editor spends ten minutes correcting a one-minute clip. Enterprise teams pay for approved output, not raw characters.
The metric therefore separates direct provider spend from human time. Teams can later apply their own hourly labor rate without VoicePilot pretending every company has the same staffing cost.
Approval gate
An observation only enters this benchmark when the test case is reviewed using the same benchmark pack and marked approved or failed under the same criteria. Provider marketing claims do not enter this calculation.
Where audio cannot be captured and reviewed, the run remains partial evidence and contributes no cost-to-approved score.
Metrics captured per test case
- Input characters and generated audio duration
- Direct generation/API cost
- Revision attempts
- Failed generations
- Human review minutes
- Human editing minutes
- Workarounds or spelling hacks
- Approved / not approved