The LiveScript blog
Real-time transcription, spoken-keyword monitoring, and what it takes to put a live broadcast in front of a person or an AI assistant while it is still airing.
How to get a Twitch VOD transcript and search what a streamer said
Twitch doesn't caption past broadcasts, so finding one sentence in an eight-hour VOD means scrubbing by ear. How to transcribe just the range you need from a VOD link, search every VOD you've transcribed at once, and keep the text after Twitch deletes the video.
How to transcribe a live stream in any language
Transcript tools assume the audio is English, and when it isn't the failure is silent - a page of English words nobody said. How the English-only checkbox and the EN / Multi badge work, why English-only is still the default, and what switching mid-broadcast does to the lines you already have.
You joined at 20:18. The stream started at 19:00.
A live transcript begins when something starts listening, not when the broadcast starts - so arriving late leaves a hole above your first line. Why that hole is usually recoverable from the broadcast's own audio, and what the repair actually does.
The stream started. You were not at the desk.
Every guide to transcribing a live stream starts with 'paste the URL', which assumes you knew it began - and for unscheduled broadcasts you usually don't. How watching the channel instead of the video gets the transcript running before you open the email.
Trading mention markets before the room hears it
Mention markets on Kalshi and Polymarket settle on a single spoken word, and everyone watches the same feed - so the edge is reaction time, not opinion. How the resolution rules differ between venues, and how to close the gap between a word being said and your order landing.
How to get a transcript of a website that isn't YouTube
Every transcript tool wants a YouTube link, but the audio you need is often on an earnings webcast, a webinar, or a council stream with no URL to paste. Why that gap exists, and how the browser extension closes it by transcribing the audio already playing in your tab.
Transcripts that know who said what
A two-person interview used to transcribe as one undifferentiated block of text. Uploads are now separated by voice automatically - what diarization is, why the transcript is built from word-level speaker tags instead of utterances, and how the names follow into SRT and WebVTT.
How to get a transcript of a YouTube stream that's still live
Every YouTube transcript tool fails on a stream that is still airing, and it is not a bug - the caption file they fetch does not exist yet. Why that is, and how to get a real-time searchable transcript of a broadcast that has not finished.