On-device transcription
Audio is difficult to search. The phrase you remember may be hidden forty minutes into a rehearsal.
Transcription solves that. Sending the recording away creates another problem.
So RECN does the work on the device. A small live pass gives words back while you record. A more careful pass follows when the take is safe. The sound does not need to visit a server first.
This took more than adding a model. The engine has to stay warm, decode the real file the recorder creates, work without interrupting capture, and fail honestly when it cannot continue.
The result is straightforward: speak, record, and search the words later.