How Crate Works
A practical look at how speech-to-text models run client-side in the browser and format subtitle files for video editors.
Audio Extraction in the Browser
When you select a video (MP4, MOV) or audio file, Crate decodes the sound locally using the HTML5 Web Audio API at 16kHz. No video bytes ever travel across the internet.
Speech Engine on WebAssembly
Using lightweight neural acoustic runtimes compiled to WebAssembly, speech recognition executes directly inside a background browser worker utilizing your device's own hardware.
Grammatical Micro-Phrasing
Raw transcripts are chunked into fast-paced 2-to-4 word phrases aligned along natural grammatical boundaries—splitting right before prepositions and conjunctions for maximum reading retention.
Contextual Visual Cues
Each phrase is analyzed in context to automatically place a relevant, high-contrast visual emoji at the end of the line, creating engaging captions formatted for viewer retention.
CapCut-Ready .SRT Export
Instant download of standard SubRip (.SRT) files with accurate timestamps formatted down to milliseconds, compatible with CapCut, Adobe Premiere Pro, Final Cut, and DaVinci Resolve.
Why Client-Side Audio Processing Matters
Most video captioning software charges steep fees because running speech models on cloud GPUs costs money with every upload. By compiling model runtimes to WebAssembly, Crate runs the speech recognition locally using your device's own hardware.
This means you do not need to wait in server queues, there is no file upload cap, and private footage remains strictly on your personal device.