Corvin 1.5.8

Запись звонков больше не отключает ваш микрофон в звонке. Раньше, как только начиналась запись, собеседники в Telegram и других приложениях переставали вас слышать: Corvin включал на микрофоне собственное эхоподавление, и оно конфликтовало с эхоподавлением самого приложения звонков. Теперь микрофон записывается как есть, а эхо из колонок по-прежнему убирается из расшифровки.

Corvin 1.5.7

Синхронизация словаря диктовки между Mac и iPhone. В настройках словаря на Mac нажмите «Подключить iPhone…» и отсканируйте QR-код камерой iPhone. Дальше слова, сохранённые на одном устройстве, появляются на другом. Словарь передаётся напрямую по локальной сети, в зашифрованном виде, без облака; при расхождении побеждает последнее сохранение.

Corvin 1.5.6

Словарь диктовки: кнопка «Сохранить» и отметка «Сохранено»; на iOS добавленные слова больше не теряются при повторном открытии экрана. На macOS поле словаря занимает всю ширину окна.

Corvin 1.5.5

Dictation can now use a dictionary of its own: names, jargon and words with an unusual spelling that fn dictation should write exactly as you list them. Find it in Settings under Dictation dictionary. Type one word or phrase per line, or load the list from a text file and save it back to one. Lines starting with # are notes for you; the model never sees them.

The dictionary is off until you turn it on. It starts with an example of three Russian words (Корвин, деплой, вайбкодинг) and short instructions. Put the most important words first: if the list is too long for the model, the words at the end are left out, and the count below the editor shows how many fit. The dictionary works with Whisper models; GigaAM does not use one, and the section says so.

Your file-transcription dictionaries are separate and work as before.

Corvin 1.5.4

Corvin can now recognise Russian with GigaAM v3, a model from Sber made for Russian. In the tests Sber published, it made about a third as many mistakes on Russian speech as whisper large-v3. It adds punctuation and writes numbers as digits by itself. Find it under Models as gigaam-v3-q8 (274 MB) or gigaam-v3-q4 (184 MB, slightly less accurate). It needs an Apple Silicon Mac and runs entirely on your Mac.

GigaAM understands only Russian, and its row in Models says so. It does not use your dictionary, and the Files card tells you when a dictionary will be ignored. With realtime dictation turned on, a GigaAM model inserts the text when you release the key instead of typing as you speak.

The speaker-recognition model (NVIDIA Nemotron 3) is now listed under Models too, with its own Download button.

Corvin 1.5.3

Transcribing "by speaker" now uses a newer model, NVIDIA's Nemotron 3 Diarization. It tells apart up to eight voices and still runs entirely on your Mac. The first time you transcribe by speaker after updating, Corvin offers to download the new model (about 190 MB). The first run after the download takes about a minute longer while macOS prepares the model; later runs are quick.

⌘, and Corvin → Settings… now open the real Settings window. Before, they opened an empty window that stayed on screen after Settings was closed. The menu bar now also has Edit and Window menus while Settings is open, so ⌘C, ⌘V, ⌘Z, ⌘A and ⌘W work there.

Corvin 1.5.2

One Files tab now holds what Transcription and History were split across: the files you add, the folder and dictionary above them, and each run's transcript.

The menubar raven widens its eye while it transcribes and settles back into feathers when it is done, so a running job is visible at a glance.

Checking for updates from the menu always answers now — up to date, the version it found, or why the check could not run — instead of leaving the menu item looking like a dead button.

Corvin 1.5.1

No changes to the app since 1.5.0. From this version on, every release has its own page with notes, listed newest first, and Corvin updates itself to it as before.

Corvin 1.5.0

Record a call from any app — Telegram, Zoom, a browser — into a transcript that alternates "Me" and "Other" replies, each stamped with its own time. Recordings are saved in parts as they go, so an interrupted call loses at most a few minutes.

Settings now live in one Settings tab; History lists recordings and transcribed files, with the app a call came from, Save as…, Show in Finder and re-transcription with any model. Cleanup of old call audio, transcripts and dictation history can be switched on, separately.

The menubar icon is a raven that opens its beak while it listens.

Corvin 1.4.0

Dictation can type text while you are still holding the key: turn on "Type text while speaking" in General settings.