TTS RECORDING SCOPE AND FIRST-BATCH ACCEPTANCE WORKSHEET
Spirelight | 17 October 2026

A blank planning aid for a custom speech collection enquiry.
This is not a dataset, recording sample, contract, quote or promise of
available speakers, facilities, rights, delivery dates or model performance.
Mark unknowns as questions. Requested specifications are not confirmed facts.

1. INTENDED MODEL AND USE
Model/recipe name and version:
Training, fine-tuning or another intended use:
Intended synthesis application and deployment:
Single voice, multiple voices or expressive/style-conditioned design:
Language, locale, accent and pronunciation requirements:
Who will approve the voice/style design?

2. VOICES, MATERIAL AND PERMISSIONS
Requested speaker profiles and voice identifiers:
Proposed speaker availability and evidence still to confirm:
Script ownership/source and permitted uses:
Required speaker permissions and intended synthesis uses to assess:
Required privacy, provenance, retention, security and transfer records:
Open rights or contract questions for the appropriate reviewer:

Do not assume ordinary recording permission includes voice synthesis,
cloning, commercial reuse or a particular deployment. Confirm the actual
terms and applicable review for the source, speaker and use.

3. RECORDING AND SCRIPT DESIGN
Required capture format, channels and original sample rate:
Target preprocessing/resampling recipe, if any:
Room, microphone, gain, session and speaking-direction requirements:
Requested styles and how intentional variation will be labelled:
Phonetic, pronunciation, prosodic and vocabulary coverage to test:
Numbers, dates, abbreviations, names, pauses and correction policy:
Missing coverage or feasibility questions:

No universal sample rate, hour count or microphone setup is prescribed here.
Agree requirements against the target recipe and inspect original capture.

4. TEXT AND DELIVERY REPRESENTATION
Required literal and/or normalized text fields:
How spoken departures from the script will be handled:
Audio-file naming and path conventions:
Duration unit and how durations will be measured:
Speaker/style identifiers and metadata definitions:
Required manifest fields and data types:
Training/validation/test split rules and separate manifests where required:
Actual loader/configuration used to check a first delivery:

Original written example only; no audio accompanies this worksheet:
Script: Dr. Lee paid $12.50 on May 4.
One chosen reading: Doctor Lee paid twelve dollars and fifty cents on May fourth.

This choice is illustrative, not a universal English normalization rule.
Record the agreed policy so a reviewer does not have to guess. A model may
expect graphemes, phonemes or another recipe-specific representation.

5. FIRST-BATCH ACCEPTANCE
Representative voices, styles and script cases the first batch must cover:
Who reviews complete utterances and compares text with the recording?
Capture checks and the agreed method/threshold for each:
Manifest/file-loading checks, required fields and unit validation:
Speaker/style-label and split checks:
How omissions, additions, wrong readings or recording faults are recorded:
Correction, re-recording, escalation and acceptance process:
Evidence and unresolved questions required before scaling collection:

File import is one check; it does not establish recording or synthesis quality.
Acceptance criteria and any model evaluation belong in the actual project plan.

6. COMMERCIAL QUESTIONS TO RESOLVE
Requested volume and how it will be measured:
Pricing unit, included preparation/review and treatment of retakes:
Required deliverables and permitted evidence-sharing process:
Target schedule, dependencies and feasibility questions:
Who confirms the final scope, rights, acceptance and price in writing?

Guide:
https://www.spirelight.ai/guides/tts-training-data
Custom TTS enquiry:
https://www.spirelight.ai/contact?view=quote&offer=tts-data-collection

Technical reference, checked 17 October 2026:
NVIDIA NeMo Data Preprocessing
https://docs.nvidia.com/nemo-framework/user-guide/latest/nemotoolkit/tts/datasets.html
Its manifest is one implementation example, not a required schema for all TTS.