DO NOT discuss the source material beyond this episode. If you want to discuss future events or theories, please use separate threads. DO NOT ask where to watch/download this episode or give links to copyrighted, non-fair use material. DO NOT troll/bait/harass/abuse other users for liking or disliking the series/characters. DO read the Anime Discussion Rules and Site & Forum Guidelines.
Good premise: Kiko is a preschool prodigy with a high IQ who is also clumsy and eccentric, similar to Kanzaki Urumi from GTO but more mature and kind rather than mischievous and malicious.
It's funny how the teacher signed her resignation papers over carrots.
The settings I used for Whisper were Large-V2 for the model, unfiltered audio (I have Spleeter on the same notebook, but I didn't use it here), language set to Japanese, init-prompt was "この作品は1996年のコメディ/アドベンチャー/日常系アニメです。主人公はキコちゃんで、先生は延長先生です。" (Made by asking ChatGPT to translate "This is a comedy/adventure/Slice-of-life anime from 1996. The main character is Kiko-chan, and the teacher is Enchou-sensei." to Japanese), task "transcribe", "VAD - Max Merge Size (s)" set to "60", "Temperature increment on fallback" set to 0.1, and "Condition on previous text" unticked.
For ChatGPT, I first split everything into (Less than, or equal to) 2048 token chunks, using the OpenAI Tokenizer, with the first prompt being "Here are some (imperfect) machine transcribed subtitles for Kiko-chan Smile EP1, please translate them to English, preserving honorifics. PART 1" (And as much of the .srt as you can fit into 2048 tokens), then for every part after, change that first bit to "Part X to translate, again, preserve honorifics.", or "Part X to translate to English, again, preserve honorifics." if it gets wrongly spits out the input text untranslated. Take note of every point where you need to click "Continue" on ChatGPT, as those lines will be a wee bit mangled, but just do each chunk, copy each one (Removing any beginning bit added by ChatGPT, if it does add something), paste it into a text file sequentially, then there you go, sometimes imperfect machine made fansubs! There are two points where the subs drop out or desync temporarily, which is a Whisper problem. For onscreen text, if you wanted to, you could use something like Google Lens to OCR it (In Google, just drag and drop a screenshot with the text clearly visible into the search bar, and you're in Lens), then have ChatGPT translate that for a good quality translation, though I didn't bother with that here.
The "raws" these were made for are the ones that have Chinese hardsubs, and Russian soft-subs, so if you have those raws, these should line up perfectly.
Fun first episode. I don't know how I found this anime, but I've had it on my plan to watch list for a while, though I've moved it up on my list of priorities because I saw a gif of it on Twitter somewhat recently. Not a 10/10 first episode, but it was light and fun.
Daviljoe193Jun 8, 2023 1:38 AM
Opinions are opinions, and differing opinions can co-exist. 🙂 But I don't do debate, life's too short to spend time debating people.
It’s time to ditch the text file.
Keep track of your anime easily by creating your own list.