Every transcription tool lets you discover after the GPU hours that your recording
was unusable. This page runs the acoustic preflight step of
TranscrIA
— a self-hosted meeting transcription portal — and gives you the verdict first:
signal metrics, DNSMOS
perceptual scores, and a difficulty timeline.
🔒 Your audio never leaves your browser.
Decoding, metrics and the DNSMOS neural model (WebAssembly) all run locally on this page —
no upload, no server, no API. (L'audio ne quitte jamais votre navigateur.)
Drop a recording here or click to choose a file wav, mp3, m4a, ogg, flac… · up to 15 minutes
Examples:
clear
watch
degraded
Detailed metrics
Metric
Value
Reading
Raw JSON (what TranscrIA's pipeline consumes)
🎧 Now hear it transcribed — in your browser too
One step further: the exact
Kroko-ASR
community model that TranscrIA ships as its zero-VRAM engine — the one that
lands near the big GPU models in our
benchmark on real meetings — running right here on one
CPU core, via sherpa-onnx
compiled to WebAssembly. Still nothing uploaded.