v0.1.2

• Auto-oto: preview additions, replacements and removals before applying; select individual changes, review alias conflicts and undo the result in one step.
• Auto-oto profiles: saved settings, prefix/suffix and separate marker corrections for initial, CV, VC, VV, VCV and ending entries; original defaults remain available.
• CVVC: VV transitions are generated alongside standalone vowels, without a separate VCV pass.
• Aliases: kana, romaji or both; bulk conversion with optional copies to keep the original aliases.
• Japanese alias coverage: missing and duplicate aliases, missing recordings and invalid timings; OpenUtau JA VCV/CVVC legacy conventions with fallback checks. This is a static check, without running the phonemizer.
• Oto playback: listen as a note (J) through WORLD, the DiffSinger vocoder or an UTAU resampler; audition transitions between two entries, with A/B comparison of original and edited markers. Transition playback requires an updated mVocalToolkit.
• Saving: preserve pending edits across multiple files and unfinished text input; failed saves keep changes available for retry. Recording and karaoke wait for pending audio saves when closing.
• Project opening: background folder scanning, loading indicator and correct selection of dropped recordings.
• macOS: reworked file dialogs, microphone permission description, bundled legacy encodings and safer quit handling.
• iOS: improved audio sessions, sample-rate conversion, interruption handling, audio decoding, file writes and background saving; local-network toolkit connections.
• UI: themed checkboxes and selection marks, clearer progress messages and translations for the new features.
• App identifier changed to pw.msan.mlabeler on Android and Apple platforms. On Android this installs as a separate app; previous settings and permissions are not transferred automatically.

v0.1.1

Quick labelling: the built-in MCFL model (13 languages) right in the app, no toolkit, on computers (Windows 7 included) and phones; language picker next to the button.
Boundary refinement now uses the built-in model, no toolkit; removed from auto-labeling.
Auto-labeling: multilingual WFL-ASR models (several languages in one song), model authors, filter by model kind, Whisper recognition errors shown in the window.
Auto-labeling model comparison: PER, boundary errors, speed, charts; without phonemes, with the labels' phonemes or both; aligners via Whisper + G2P.
Unsure phoneme check with percent and threshold (off by default); a manual edit clears the mark.
Sound editing: spectrogram brush with a soft edge.
Dataset summary: phoneme chart.
Settings: all sections expanded, Expand all / Collapse all.
Phones: scale button on by default, pinch no longer grabs boundaries.

v0.1.1-beta2

Auto-labeling: Whisper by default with model choice, breaths (AP), segment-wise re-alignment of long recordings, TIFA, text and unknown-word check before running.
Audio: save with drawn pitch, V/UV and volume pencil, gain change in dB.
Formats: label conversion (lab, TextGrid, Audacity, .ds, oto ⇄ .inf), .ust import, NiaoNiao voicebanks creation, Mac filename normalization.
oto: jump to the next recording after a manual edit, compare with another oto.ini.
Plugins on multiple files with preview, extended API.
Labeling time tracking and Discord Rich Presence.
UI: animations, collapsible settings sections, Mintaka theme, faster background image, “?” hints.
Auto-update on Windows and Android.

v0.1.1-beta1

Interface: the menus can go in the window title (on by default, switchable). Themes choose the title’s look, and Retro gets a blue one with round buttons. There’s a map of the whole recording that also works as the scroll bar: drag to move, drag an edge to zoom. Animations are smooth and can be turned off. Phonemes can be tinted by kind. Themes can have a background gradient or picture. Long lists can be scrolled by dragging the bar anywhere at once.
File list: right-click menu (rename, delete, show in folder). Several recordings can be picked to join them into one or delete them together.
Boundary refinement with a new refiner model (mRefinerModel), off by default.
Automatic oto:
more accurate vowel starts and syllable finding;
ARPAsing, pinyin and CVC styles;
kana aliases for romaji banks
Karaoke: music volume, and a choice of voice separation and recognition models.
Tools: cut out phonemes into separate files.
Checks: a pause at the start and end of a recording can be required.
Tip of the day.
Windows:
fully portable build;
fixed start with non-Latin user names;
experimental support for Windows 7 SP1 and 8.1 (without the toolkit).
Toolkit: updates itself even when it was already running. Added a Reinstall button.

v0.1.0

README: screenshot