I’m at the point where I’m trying to watch as much as possible with Japanese subtitles. You may be in a similar boat. I’ve been using resources like kitsunekko.net and Netflix Japan (with VPN) but Netflix Japan doesn’t have everything, and most subtitles on kitsunekko.net also require you to find the exact rip that they used which is usually bad quality or suffer by trying to fix mis-timed subtitles.
Fixing the timing of subtitles is arduous, even with automatic sub re-timing tools I’ve never had success but recently I had the idea, what if I generated my own subtitles with AI? Voice actors have very good pronunciation so this is a perfect use case for AI. Using MacWhisper I run it on-device so it’s very fast and free (I generated subs for an entire season of anime in less than 30min) and after watching 5-6 episodes with my wife (Japanese), we’re very impressed with it.
A few caveats to know- one, it’s bad at kanji for names. If you can’t yet understand names from hearing them it’s probably too soon to try this out. We also noticed it would sometimes get the kanji wrong for certain fantasy words but if you have a dictionary handy like Shirabe Jisho it’s easy enough to look up and figure it out and regardless, it’s way better than having no subtitles at all or English subs for learning.
If anyone is interested in a full tutorial you can run yourself, let me know. For MKV files I needed to extract the audio because MacWhisper doesn’t support the format so I wrote a quick script that extracts the audio with ffmpeg very quickly and you can drag that to MacWhisper to batch generate the subs (SRT) from it.
I have a couple of older Intel-based MacBook pros - one of them a mid-2015 running whatever is the last macos version that it supports (Monterey?), and a second one that IIRC is 2019 vintage, maybe running Sonoma (?).
I think MacWhisper will be slower than what I said above on the Intel Macs because their AI prowess isn’t as high as the M series chips, but definitely feel free to give it a try and let us know how it goes. Here’s the script I made to extract audio from MKVs, then just drag the MP4 (audio-only) to MacWhisper to generate the SRT subtitles.
# Run ffmpeg -i FILENAME.mkv first on one of the episodes and copy the Japanese audio channel number and replace `-map 0:2` with `-map 0:X` with X being the number you found.
find . -name '*.mkv' -exec ffmpeg -i {} -map 0:2 -vn -c copy {}.mp4 \; ; echo 'ffmpeg finished'
I do have to take a few steps back, though - while I have used ffmpeg to do various ‘unnatural acts’ with video (in a different lifetime - or was it an alternate timeline? I forget…), I don’t yet have any suitable video source(s) on hand to play with (yet).
But if it looks like it’s promising, I might consider getting one of the new Mac Minis which I could dedicate to that purpose.
I first tested it on some downloaded YouTube videos and compared to YouTube’s automatic captioning and Whisper crushed it - I thought “wow I can probably just watch along with these subtitles” whereas YouTube’s automatic captioning is a struggle because of the constant mistakes it makes. I think you’ll be impressed, you can even just drop a YouTube link in it below and it will download it for you (also, I used the Large v3 Turbo model for all my transcriptions and manually set the language, if that helps anyone else trying) -
