lexic alpha connecting…
Press to start talking
Tap to start talking
Hold space to talk

What this is

You talk. It writes the note. The listening happens on this server, not on someone else's — and it can be taught the words your work actually uses.

What happens when you speak

01Your browser captures audio Resampled to 16 kHz mono and streamed as it is spoken. your browser
02Silence is cut out A voice-activity detector finds where speech starts and stops, so only real speech is transcribed. this server
03Speech becomes text A speech model runs on this server's own CPU. No transcription service is called, and the audio is not sent anywhere. this server
04Domain words are restored Jargon the recogniser mis-heard is matched by sound and corrected. Every change is shown to you. this server
05The text becomes a note Title, summary, tags and action items. This is the only step that leaves the server. a hosted model

Why domains matter

Speech recognisers fail on jargon in a particular way: they produce something that sounds right and is spelled wrong. A product name comes back as a near-homophone; a drug name becomes an everyday word.

Choosing a domain tells the system which words are likely here. In a dealership, “highlights” is probably “Hilux”. In a photo studio it is not. A single global word list cannot hold both preferences at once — which is why the vocabulary is per-domain rather than one big dictionary.

Corrections only apply where a word genuinely sounds like a domain term, and each one is listed under the note so you can see what changed.

This instance

Speech recognition
Note writing
Where your words go
Your recording
Longest recording

Lexic needs your microphone

Release space or click to finish
Writing your note

What you said

Drop to transcribe