If live transcription has to stop during a take, the record screen now tells you. The recording carries on, and you can transcribe it when you stop. Stability improvements.

Turn on Transcribe While Recording and follow the transcription as it is produced. The words appear on screen as they are transcribed, a little behind what you are saying, and the first ones arrive within seconds. Sharing now asks what to send: the audio, the transcript as text, or as a .md file. Stability improvements.

• Pinch the waveform to zoom in and find a moment precisely. • A chunk that fails to transcribe no longer leaves a silent gap. DiPLO retries, and if it still fails it stops where it was, so you can continue from that point instead of being handed a transcript with a hole in it. • Speaker identification no longer stalls when a window fails: the recording comes back without labels rather than with a gap in them. • A clearer message when the device does not have enough free memory for the model.

• More natural transcripts: paragraphs grouped better, fewer sentences cut in half and fewer words dropped. • Better speaker identification: it tells people apart more reliably and no longer swallows whoever speaks least. • Fixed an unexpected quit when starting a recording and when playing back at 1.5× and 2×.

Import audio you already have: from Files, from a video in Photos, or shared straight from another app. Tapping a search result now starts playback at that word. Plus stability fixes.