# p05 Try again. — local voice test v2 assembly QA

- Time: 2026-05-25 20:37 CST
- Actor: @組裝Agent / @聲音Agent / @審核Agent
- Artifact: `assembled_segments/p05_try_again/p05_try_again_3pass_local_voice_test_v2.mp4`
- Source clip: `source_clips/p05_try_again/p05_try_again_v5_8s_trimmed_visual_only.mp4`
- Local voice: `audio/local_f5tts/p05_try_again_B_childlike_slow075.mp3`

## Assembly

3-pass teaching structure:

1. Pass 1: no subtitles + local F5-TTS `Try again.` at the encouragement beat.
2. Pass 2: English subtitle `Try again.` + same local voice.
3. Pass 3: English subtitle `Try again.` + Traditional Chinese `再試一次。` + same local voice.

Voice timing: around 4.55s inside each 8s pass, matching Mia's encouragement/rebuild beat.

## MP4 verification

- Duration: 24.023s
- Resolution: 1280×720
- FPS: 24
- Video: H.264, yuv420p
- Audio: AAC, 44.1kHz, mono
- Size: 5,526,018 bytes

## STT QA

Whole-file STT over long silent sections merged/repeated segments imperfectly, so each expected voice window was extracted and checked separately.

- Pass 1 window 4.2–6.4s: `Try again.` — pass exact
- Pass 2 window 12.2–14.4s: `Try again.` — pass exact
- Pass 3 window 20.2–22.4s: `Try again.` — pass exact

## Subtitle visual QA

Contact sheet: `qa/assembled_contact_sheets/p05_local_voice_test_v2_contact_sheet.jpg`

Vision check:

- First sample frame: no subtitles.
- Second sample frame: readable English `Try again.`.
- Third sample frame: readable English `Try again.` and Traditional Chinese `再試一次。`.

## Verdict

`test_ready_awaiting_human_voice_review` — score 8.0/10.

Caveats:

- This is a local voice test, not a provider-voice final source.
- User should judge whether the local F5-TTS voice is childlike/natural enough.
- v5 visual source still has the prior caveat: failure collapse moment is partly inferred from before/after states.