Changelog: release notes

Newest first. Every earlier release, with installers and tags, is on GitHub Releases.

v0.9.42026-11

Punch in over a take, split a mix into stems, and import M4A.

New: Punch replaces one region of a track with a new take while you hear the rest of the mix; you set the delay for your speakers and microphone once in Settings → Audio. separate_stems splits a track into vocals, drums, bass and other on this computer, after you download the stem model (about 171 MB) in Settings → Account.
New: M4A/AAC files import on macOS and Windows, through the system's own decoder. render_final exports AIFF and OGG (Opus), and the whole session or just a range of it. export_transcript writes SRT, WebVTT or text, and transcript_to_markers adds timeline markers. New tools: find_markers, add_markers, edit_markers, delete_markers, join_clips, align_end_to_end, reorder_tracks, mute_all_tracks and wah_wah; fade curves, shelving EQ bands, multi-tap echo, and snap-to-zero-crossing on the cut tools.
Changed: recording delays you measured or typed in an earlier version read as not measured, because the audio library was updated (cpal 0.18); measure again in Settings → Audio before you punch in. Fixed: a punch-in take is applied only to the project it was recorded in, never to a copy of it; a track's transcript moves with the track on time_shift and move_clip, exports are refused inside the project's own folders, Save As no longer calls a copy working when a generated tone is missing from it, and Escape closes the template picker.

Measure loudness, find who speaks when, and import more formats.

New: measure reports loudness, peaks, noise floor and clipping, with an ACX check. diarize finds who speaks when, on this computer, after you download its 33 MB speaker model in Settings → Account.
New: generate_click_track, generate_chirp and generate_dtmf make click tracks, sweeps and keypad tones. AIFF and Opus (mono and stereo) files import.
Fixed: the loudness-master skill no longer raises gain after the limiter.

One keychain entry for your keys, and transcription that no longer skips speech.

Changed: keys and the ChatGPT sign-in share one keychain entry, so macOS asks for your password at most once per launch. Existing keys move over on the first launch.
Fixed: transcription no longer misses stretches of speech, and word times are closer.
Fixed: Retry sends your message once, empty tracks get a lane, and tone and noise work in an empty project.