Speech and language sampling
Language sampling, with the recording still attached.
Arc is an editor for clinical speech samples. Record a child, get a draft transcript, and check it against the audio word by word — fast enough to actually do it.
A synthetic sample: two voices reading a scripted retell with planted child forms. No real child was recorded. This is a frozen slice of the editor, not the app — marks you make here go nowhere.
The transcript is the hard part.
A language sample is some of the best evidence a speech-language pathologist can put in a report, and most of them almost never collect one. Not because the analysis is hard — mean length of utterance, words per sentence, percent consonants correct are arithmetic once the transcript exists. It's that getting from a recording to a verbatim, speaker-attributed, utterance-segmented transcript means hours in a Word document with an audio player, and nobody has hours.
Consumer transcription doesn't help. It's built to delete the "um," collapse the false start, and quietly turn goed into went. In clinical speech, those are the data.
So Arc is built around the correcting.
Arc starts from an automatic draft and gives you an instrument for verifying it. Every word is bound to its moment in the audio: click to hear it, drag to retime it, split or merge lines, swap a speaker, mark fillers, retraces, and errors. Where the recording has speech the transcript doesn't account for, Arc flags it and lets you listen.
When the transcript is verified, the measures compute themselves and the report is ready to sign. The draft is the door; the editor is the room.
Where it is
In development. Not yet available.
Arc is being built now, and a small alpha is coming. If you'd like to be in it — or just want to know when it's ready — leave your email. No newsletter, no drip campaign; you'll hear from a person when there's something to show.