2026-03-14

Testing YouTube AI Voice 1

What I am currently doing
Rt2112 reading x BGM
English-based localization for the world → auto-dubbing
Capturing demand for mindfulness and business hacks in the work/sleep BGM genre

Today's results

### Auto-dubbing fully verified

- Vol. 11 (Vol. 009, Episodes 1-4) Fish audio x 58 minutes x 1 mixed track, 20-language auto-dubbing confirmed
- Standard TTS languages (AR, NL, HE, JA, KO, ML, PL, RU, UK) published same-day
- Expressive Speech target languages (ES, PT, DE, FR, IT, HI, ID) take 2-3 days to process (due to heavy AI processing)
- Expressive Speech = High-quality dubbing with emotion and intonation. All main target languages are included

### Cost finalized

v5 (Old)

- ElevenLabs: $99/month (¥14,850)
- Fish Audio API: $0
- Suno: ¥1,500
- Total: ¥16,350

v6 (Current)

- ElevenLabs: $0 (Cancelled)
- Fish Audio API: ¥515/volume (2 volumes/month ≈ ¥1,030)
- Suno: ¥1,500
- Total: ¥2,530 (85% reduction)

### rt2112_voice.py v6 complete

- No SDK dependency (direct requests HTTP calls. Python 3.9 compatible)
- Automatically removes ◇, ———, and ## headings before sending to Fish API
- Manuscript files are not changed at all (read-only)
- ¥515 for 28 episodes. Cost can be verified via dry-run
- Setup: pip install requests + export FISH_API_KEY=key

### Positioning finalized

- Channel name: Road to 2112: Cozy Sci-Fi Relax Stories
- Sleep Stories → Cozy Sci-Fi Relax Stories (foregrounding work-focused reading BGM)
- Description template v6 version created
- Added language learning use case (excellent compatibility with 20-language auto-dubbing)

### Posting operations

- 3-4 videos per week (starting at a pace of 8 per month, keeping a backlog)
- Posted at 23:00 JST (= 12:00 PM NY time)
- Scheduled post: Set to public 3 days after upload (waiting for auto-dubbing to complete)

## BGM dB test in progress

1st video (Vol. 11): -12dB / Fish / Standard languages published. Expressive processing
2nd video: -15dB / Fish / In progress
3rd video: -12dB / Fish / Uploaded
Test (unlisted): -20→-15dB / Fish / Under verification

### Findings

- At -12dB, the dubbed audio becomes too strong after auto-dubbing, causing the BGM to fade out
- -15dB provides a more natural balance
- Due to YouTube's audio separation and re-mixing mechanism, the AI performs better with lower BGM levels
- Standard BGM dB value to be finalized after test results

## KDP

- Successfully generated all 14 EPUBs for Vol. 008 (7 languages x blue/yellow)
- KPF not outputting → Kindle Previewer was uninstalled. Re-installation required

## Next steps

- Wait for Expressive Speech languages (ES/PT/DE/FR) to be published → Quality check
- Finalize standard value based on BGM dB test results
- Re-install Kindle Previewer → Generate Vol. 008 KPF
- Start mass production of Vol. 009, Episode 5 and beyond
- Test upload 4th video at -20dB

## Finalized production flow (v6)

1. Manuscript: md (as is, do not touch)
1. Audio: rt2112_voice.py → Fish API → mp3 (auto-removes ◇/———/##)
1. Editing: Proofreading in Premiere (fixing line breaks/howling) + BGM adjustment
1. Video: Gemini/Veo + Premiere
1. Upload: YouTube (Language setting: English) → Scheduled public release 3 days later (23:00 JST)
1. Multilingual: Auto-dubbing (20 languages, $0)
1. KDP: mk_all_epub.sh → EPUB → KPF → KDP publishing

## Tool list

- rt2112_voice.py: Fish API audio generation (direct HTTP) ✅ v6 complete
- mix60.py: 58-minute audio mixer (ffmpeg) ✅ Complete. Manual Premiere work is primary
- mk_all_epub.sh: Batch generation of all language EPUBs ✅ Operation confirmed
- mk_epub.py: EPUB/KPF generation ✅ KPF waiting for Previewer re-installation
- kdp_toc.py: KDP table of contents list generation ✅ Operation confirmed

## Goals

- ¥500,000/month (must achieve): 670,000 monthly views. Within 1 year
- ¥1,000,000/month (stretch): 1.33 million monthly views
- Monetization (YPP): Expected to reach with 30-50 videos. 3-4 months