CUSTOM VOICE READINESS WORKSHEET
Soniox TTS v2 / Inworld TTS-2 | Reviewed October 23, 2026

Original planning resource with unperformed fixtures. No cloning, generation,
listening study, API benchmark or accepted production run is represented here.
The worksheet is free; provider and production services can have separate costs.

1. REFERENCE AND PURPOSE
Permitted speaker / reviewer (use a nonprivate internal reference):
Permission record, intended use and limits:
Business task:
Intended audience and language:
Original recording reference / version / date:
Recording quality concerns:
No private caller/customer recording or another model's generated output.

2. PREPARATION RECORD, ONE PER CANDIDATE
Candidate label (A/B, hide provider from listeners):
Provider / account scope:
Exact model ID / provider lifecycle checked on:
Reference version:
Preparation route:
Prepared voice ID (no credential):
Readiness for intended model / observed evidence:
Preparation date and procedure used:
Error / unresolved state / corrective action:
Eligible to proceed? YES / NO / UNRESOLVED

Use the current provider procedure; do not assume an upload or saved identifier
establishes readiness for every model, project or workspace.

3. FREEZE THE EVALUATION BEFORE GENERATION
Script version:
Output format / rate / playback path:
Client location / timing boundaries:
Number of attempts planned per script:
Permitted reference audio used consistently:
Critical content errors that fail acceptance:
Identity review criteria and authorized reviewers:
Other changing variables and why:
Approval to incur provider costs obtained from:

4. ORIGINAL ENGLISH FIXTURES
S1 - service information
The workshop enquiry is a request. A team member will confirm the available
dates. What would you like the team to know?

S2 - correction and identifier
The reference is B7, not D7. The entrance is beside the garden gate.
Would you like me to repeat the reference?

S3 - longer delivery
To help the team review your enquiry, please give the workshop name and the
day you had in mind. Your preferred date is not yet reserved. If the plan
changes, the team will explain the options before confirming anything.
Which workshop are you asking about?

5. OBSERVATION ROW, REPEAT FOR EACH ATTEMPT
Blind label / script / attempt:
Provider acceptance evidence:
Important words intelligible and unchanged? Evidence:
Correction, letters or numbers correct? Evidence:
Speaker identity resemblance (separate from pleasantness):
Naturalness / delivery comments:
First audible speech timing and boundary:
Unusable audio / errors / failed attempts:
Accepted? YES / NO / UNRESOLVED; reason:

6. REVIEW WITHOUT CHERRY-PICKING
Did reviewers hear the same permitted reference?
Reviewer familiarity with the intended speaker:
Which labels and details were hidden from reviewers?
What result would change the decision?
All attempts retained, including failures:
Additional language-specific review needed:
Any identity claim unsupported by this review:

7. COST AND APPLICATION READINESS
Actual provider units per attempt:
Applicable plan / tariff / effective date:
Preparation costs and staff time:
Synthesis usage cost, including discarded outputs:
Accepted output count:
Other platform / model / storage / telephone costs:
Exact application model and voice choices available:
Configuration / provider acceptance / listener review evidence kept separate:
Decision: PREPARE / RETEST / READY FOR SCOPED APPLICATION CHECK / STOP
Decision owner / date / next review trigger:

Stock-voice benchmark scores do not establish cloned-speaker resemblance.
Current provider preparation contracts:
https://soniox.com/docs/tts/concepts/voice-cloning
https://docs.inworld.ai/tts/instant-voice-cloning
https://docs.inworld.ai/tts/professional-voice-cloning
Companion article:
https://burki.dev/blog/soniox-vs-inworld-custom-voice-readiness