I've been using the Comprehensible Input method (based on Stephen Krashen's work [0]) to learn languages. A common pain point is that YouTube moves too fast, I needed time to mine words or shadow the speaker without constantly fumbling for the spacebar, I also wished that I had the translation of the sentence and maybe even insights on the why behind the grammar used in the sentence (e.g. in Japanese to express causality we can use から - kara vs ので - node and choosing the right one depends on politeness and sentence structure).

To solve this, I built Subtitle Insights [1]. It uses Chrome's Built-In AI (Gemini Nano). This means all the translation, grammar analysis, and cultural nuance explanations happen locally on-device. It's free, there's no API cost for the user and no data being sent to an external LLM server for processing because all processing happens in your device (review the hardware requirements here [2]).

Key Technical Features:

Local Inference: Uses the Gemini Nano integration in Chrome.

Auto-Pause: It pauses at the end of subtitle segments to allow for "processing time."

Customizable Prompting: You can tweak the system prompt to change how Gemini analyzes the sentences (e.g., focus on slang vs. formal grammar).

Subtitle Sync: Added a tool to manually re-sync external subtitle files with the video audio track.

Keyboard First: There are shortcuts for segment-to-segment navigation and segment replays.

[0] https://www.youtube.com/watch?v=fnUc_W3xE1w [1] https://mauriciopoppe.github.io/SubtitleInsights [2] https://developer.chrome.com/docs/ai/prompt-api#hardware-req...

Show HN: Subtitle Insights – On-device AI translation for YouTube subtitles