2026-04-09
Youtube Auto-dubbing: After 20 videos, a strategy emerges. Rt2112 Development Log | April 5 - April 9, 2026
I've been running this channel since around March.
Here is the current status.
Main channel data (as of April 9)
28 days: Views 1.7K / Impressions 34.1K / CTR 2.4% / AVD 2:52
Compared to previous 28 days: Views +751%, Impressions +862%
Subscribers: 15 (+10)
Total watch time: 62.3 hours
Monthly audience: 774 (trending upward)
50 views in the last 48 hours
Continuing test distribution for the algorithm. Impressions are stable every day
It looks disastrous at first glance, but in terms of what I'm doing, it's within expectations. I've been repeating trial and error since March, and I can see the parts that generally work, so here's that story. First, as a major premise, the negative elements of auto-dubbing must be factored in, which is as follows.
Main channel data by country (first appearance)
US: 66 views, AVD 6:33 — Expressive Speech zone. It's being listened to properly
Japan: 265 views, AVD 2:13 — Standard TTS. Listen and drop off
Ukraine: 103 views, AVD 1:07 — Standard TTS. Immediate drop off
South Korea: 40 views, AVD 0:46 — Standard TTS. Immediate drop off
The hypothesis that the standard TTS zone is dragging down the overall AVD has been confirmed
Looking only at the Expressive Speech zone (US, etc.), an AVD of 6:33 is a sufficient figure
No countermeasure. Waiting for views in the ES/PT/DE/FR zones to increase
👉 If you listen to the Japanese auto-dubbing, you'll understand immediately; it's unlistenable. Especially since it's for bedtime sleep use, the TTS voice is at a level that makes you want to say, 'Are you kidding me?' Since I'm using English as the axis and auto-translating, the text is simple and rough, so it's like, 'Well, of course you can't listen to this.' To say it first, the book on the Kindle side is translated from Japanese to English. The Japanese is quite easy to read (or should be). Since it's different from that text, well, it's unavoidable. Similarly, there is the same poor audience retention in Ukraine and South Korea. It's likely certain that the TTS is dragging down the AVD. In the end, if this metric doesn't go up, impressions won't grow (the algorithm thinks it's a video with low satisfaction), so it's a 'good luck with the updates, YouTube' situation. Since I prioritized uploading videos on the premise of updates from the beginning, it was already factored in, and on the other hand, I'm fine as long as the AVD in English-speaking regions can be maintained. Because of that, among the 10 or so videos currently scheduled, I'm moving up the release of those where Expressive Speech has been completed. I think it will change again if they target these countries.
Actually, this is likely the primary language for the YouTube management side.
(It's Indonesia rather than Japan—I suppose that makes sense. That's just how the Japanese market is.)
Well, for that reason, the view counts aren't going up at all, but there has clearly been a change in the algorithm.
Changes in initial performance for new videos on the main channel
A video with a 4.1% CTR and 4:25 AVD was released (ranked 1/10 by views).
Also a video with a 3.4% CTR and 6:33 AVD (ranked 4/10).
Signs of a shift from previous 'scattershot' tactics to 'pinpoint delivery to the target audience'.
Views are in the single digits, but retention and CTR are both showing green arrows.
The algorithm is at the stage of considering expanded delivery to similar clusters.
👉 While I am grateful that impressions have been increasing since the first month, the CTR was consistently low due to the scattershot testing; however, even though impressions aren't skyrocketing, the CTR and AVD have started to rise.
Ultimately, the CTR drops again after about 3 days of publication, which is proof that I've started to identify the core audience.
It is an optimistic hypothesis that it will start showing the content to viewers in the same cluster.
Either way, it seems important to keep producing content.
Since I have a stock of content ready from triggering the auto-dubbing process, which takes about 2 weeks, to the actual publication, I can now continue to release content consistently on the same day and time with better precision.
In addition, because the content creates stories from news, ensuring freshness is necessary.
Currently, there is about a 3-month time lag from the news used to create the video from past story stocks to the video publication.
However, since the story stock will soon run out, I will resume story generation starting next week.
Then, these will be published around mid-May, allowing me to keep the time from news to video publication within one month and align the pace.
Lo-fi 2112 Data (as of April 9)
28 days: Views 215 / Impressions 6,343 / CTR 2.0% / AVD 16:02
Compared to previous: Views +84, Impressions +2,542
The right end of the graph is rising sharply. Recently reached 30 views per day.
An AVD of 16 minutes is 27% for a 60-minute video. Sufficient for a BGM channel.
AVD from Japan is 2:16. The structure where other countries push up the average is the same as the main channel.
Cozy Cosmic Vibes is at the top with 44 views. Approaching the 100-view line.
Lo-fi 2112 DJ Character Composition Finalization
Decided on a policy to unify to a front-facing DJ booth composition.
Fixed: Front-facing DJ booth, Lo-fi 2112 logo
Variable: Player (cat robot, dog robot, bird, human, etc.), background (Distant theme)
This also serves as differentiation from the main channel (landscape composition, story-driven feel).
First draft of the ennui scarf cat robot DJ complete. Will be applied from the next video.
The Lo-fi side will follow a 'go all out with playfulness' policy. A contrast to the dignity of the main channel.
👉 This is the sub BGM channel. I have a stock of music built up, so I've been reusing it, but it's actually doing quite well.
Since I localized the titles into multiple languages last time, impressions have been increasing, so now it's just a matter of improving the CTR. That's why I'm going with the DJ composition approach. (Honestly, the numbers are down, haha).
But since it's a playground, I'll keep it going.
Lo-fi 2112 BGM startup issue
Starlight Dome Nights: AVD 0:28 (Immediate drop-off on a 56-minute video)
Cause: Ambient tracks have a slow startup at the beginning.
Countermeasure: Start the melody within the first 5 seconds. Start with a pleasant sound rather than a fade-in.
👉 I hadn't paid much attention to this, but it seems important. Ambient tracks have
about 10 seconds of 'woooooooosh' atmospheric buildup, but apparently, people drop off during that.
Need to have programmed sounds from the start.
While thinking, 'How much of a hurry are people in?', I also thought, 'I see,' so I'm trying it out just as the AI suggested. It would be interesting if it works.
Multilingual title automation successful
localize.py v2 deployed to production.
Batch settings for 25 languages completed for 'Distant City, Pre-Dawn Study'. Finished in 10 seconds.
Authentication files consolidated into config/. Directory organized.
Operational guide created. A system that I can run by myself.
👉 As mentioned earlier, this changed the growth of impressions. The good thing about the BGM channel is that it doesn't need words. In fact, since the main channel is struggling initially with language, this one might have a faster startup.
(Though the BGM genre has the problem of not gaining subscribers.)
TuneCore distribution started
Standard plan subscription (9,900 yen/year, 15 songs/album).
Artist name: Sassa Mahha (Unified under the sassamahha name. Coexistence with future AINOA, etc.)
First album 'Lo-fi2112 / Distant Workshop, Craftsman Dusk' released.
Distributed to all services including Apple Music and Spotify (Content ID only turned off).
Input semi-automated with Claude in Chrome. Only music upload is manual. About 10 minutes/album.
Tags: lo-fi / study music / música para estudiar / música para estudar / BGM for work
Linkcore album URL in the pinned comment. A complete funnel in one line
Break-even point: Recovering 9,900 yen with approximately 15,000–20,000 annual views
👉 This is also a purely hobby-based project, but I have decided to live as a self-proclaimed composer.
Linkcore and Funnel Design
Lo-fi 2112 pinned comment: One Linkcore URL per album
No need to drive traffic from the main channel to Lo-fi (risk of algorithm contamination)
Profile links: Spotify / Apple Music / YouTube / Amazon / Linkcore
Workflow v1 Formulation
Documented mass production system for Vol. 36–Vol. 100 (Kindle Vol. 10–19)
5-stage process: 1. Material creation → 2. Video creation → 3. Distribution → 4. Bookbinding → 5. Promotion
Organized dependency tree. Parallel tasks can utilize agents
5 revenue channels from a single manuscript (YouTube/TuneCore/Kindle/POD/note)
👉 Currently, I have 9 volumes in stock, which is enough for 35 YouTube videos. Once this supply runs out, I will resume story generation. Promotion (step 5) is currently left to YouTube, but I want to start using Pinterest.
AINOA Connection and Expansion of the Distant Series
Concept to expand the Distant series locations to the entire globe
Distant Istanbul, Distant Rio, Distant Kyoto, etc.
Blending traditional instruments from each country into Lo-fi (e.g., Japanese instrument Lo-fi)
Synergy with multilingual titles (Lo-fi tracks with country names in the title are easier to click)
Considering the potential for a dedicated Japanese instrument Lo-fi channel (building up a library as a hobby)
Seed AINOA city names in album and song titles. Do not include them in the channel name.
👉 It started with another silly observation, but for some reason, Lo-fi channels with a "Japanese taste for Americans" get high view counts. It's like a Zen vibe.
(It feels like they are taking the stuff that goes viral on Instagram with oversaturated colors seriously.)
If that works, I think Japanese instrument Lo-fi would be good too.
Once I enter mass production, I'll have half my Suno credits left over every month, so I want to use them for experimental slots, and maybe even do some instrumental tracks with a Chemical Brothers vibe.
Occupation: Composer (Provisional)
Strategy
Main channel: 50 videos by May. Prepare for the June update. No changes.
Lo-fi 2112: Test CTR improvements with a DJ character composition. Revise to be instant-start background music.
TuneCore: Stack albums in sync with videos. Place Linkcore links on all videos.
Side note.
Once the current flow is established, I could outsource the work until monetization, or lately, I've been wanting to secure some kind of fixed income. Since I'll have a lot of free time once I enter the mass production flow, I'm wondering what to do with that spare time.
I don't have to do it, but as you might guess, even though AI can do many things, the essence of a successful startup is "problem solving," and AI is just the means, so unless you dive into where there's smoke, you won't smell the money.
As long as I'm doing BtoC, it takes time to adapt to the platform, and it costs money to speed things up, so I'm wondering what to do with my free time.
For now, while I start thinking about such things bit by bit, I will calmly stack them up according to the workflow.