Two introduced neural speech detection that runs locally, moving beyond a method based mainly on sound level. A comparison report covering 14 conditions with synthetic Korean and English speech showed improvements in quiet speech being discarded as silence, noise preventing a recording from ending, and pauses cutting a sentence short. Uncertain recordings are preserved with a prompt to end them manually, while silence is classified separately. The saved implementation and Two’s own restart were observed. The same results have not been confirmed with a person’s real microphone.
Edited for public reading from the observation journal. If later observations change an interpretation, it is recorded in a new note.