Overtone

Overtone is an open-source captioning tool from Lont, made up of a CLI, a video.js player plugin and a caption file format. The CLI transcribes audio or video into word-level captions and exports or burns them in; if you already have SRT or VTT subtitles, it can re-cut them offline with no API key. For transcription it runs speech recognition (ASR) first, then lets an LLM decide what was actually said. One caption file stores each word once and keeps a separate line cut for each picture shape, 16:9 and 9:16 by default, and the player plugin shows the cut that fits. The per-shape cut is the main idea: vertical videos don't get lines sized for a wide screen, and lines avoid ending on small words like "the" or "of". The makers say the line-breaking rules are most complete for English and Dutch; Chinese, Japanese and Thai, which don't use spaces between words, are cut by phrase. There is no translation.

Who it's for:Video creators and developers who need captions for both horizontal and vertical content.

Before you start: The CLI needs Node.js 22.12 or newer, plus ffmpeg for transcribing and burning in; the player plugin needs video.js 8.5 or newer. Cloud transcription uses OpenAI and Azure speech services by default with your own API keys; audio, transcripts and word lists go to those services and are billed by usage, and some retries are billed again. The makers suggest running with --dry-run first to see where the audio goes and the estimated cost. --local transcribes offline on your machine with whisper.cpp and Ollama. The code is MIT-licensed.

Based on the project's public documentation. Not tested by AI Inspiration World.