# Voice-actor directions: None — Tamil

Record only text marked approved in `recording-script.json`. Do not translate, paraphrase or correct wording during the session. Stop and flag unnatural or ambiguous text for the native reviewer.

## Performance

- Speak warmly, calmly and respectfully to one learner; avoid baby voice, exaggerated cheerfulness or theatrical emotion.
- Keep normal connected-speech rhythm. Do not over-enunciate unless a separate teaching take is explicitly requested.
- Preserve the communicative function: requests should sound genuinely useful, questions should have natural intonation, and feedback should remain neutral and non-judgmental.
- Record every line as an independent utterance. Do not add the concept label, explanations, counting, clicks or slate speech.
- For `stop`, `no`, refusal and help messages, sound clear and immediately understandable without sounding angry or coercive.
- For caregiver directions, use a collaborative professional tone, not a command directed at the learner.

## Language-specific direction

Record the approved spoken Tamil variety, not an artificially literary reading unless marked. Preserve vowel length, retroflex contrasts and natural child-directed rhythm.

## Technical capture

- Quiet treated room; one consistent microphone position 15–20 cm from the mouth with pop protection.
- Mono WAV, linear PCM, 48 kHz, 24-bit; never record directly to MP3.
- Capture 250–500 ms clean room tone before and after each utterance.
- Aim for approximately -20 LUFS integrated per short utterance after non-destructive normalization; true peak must remain at or below -3 dBTP.
- No music, reverb, noise gate pumping, pitch correction, voice cloning or aggressive denoising.
- Keep raw takes immutable outside the delivery folder; export selected masters using the exact filenames in the script.
- Record at least two clean takes. Add further takes only for a documented pronunciation, pacing or noise issue.

## Session and safeguarding

- Confirm the actor release and approved language variety before recording; store releases outside this content repository and reference only their identifier.
- Do not record identifiable learners or confidential clinical material.
- Permit breaks and stop immediately on discomfort. Do not imitate a disability, accent or child voice stereotypically.
- The actor may flag culturally inappropriate or unnatural wording without being asked to improvise a replacement on mic.

## Acceptance checks

- Filename matches the manifest exactly.
- Spoken words match approved text exactly while remaining natural.
- No clipping, mouth clicks, background speech, electrical hum or abrupt cutoffs.
- Native reviewer approves pronunciation and variety.
- SLP approves clarity, pacing and communicative function.
- Editor records duration, loudness, checksum, selected take, actor release reference and review status in `assets.csv`.