Heyra 0.6.0

• Parakeet v2 joins the voice models: English, best of all on meetings and calls on
  heyra-bench, ahead of Parakeet v3 and Bark Clean word for word.
• Voice model and Settings… in the menu bar menu.
• Settings regrouped: push to talk, voice model, mic and open at login first, then
  Dictation, Output and Appearance. Plain labels.
• While dictating: do nothing, mute, or lower the volume to a third.
• The orb can sit at the top, under the menu bar; the words, toasts and copy card follow it.
• Dictionary as boxes: when I say this, write that. Tab, Enter, Esc, ⌘V; saves as you type.
• History shows the app each take went into, the model, its length and the time to text.
• Version and Check for updates at the bottom of the sidebar.
• Setup: shorter, plainer steps; it asks for Accessibility once, then opens the pane,
  and opens Home when it's done.
• New installs start on Bark Clean even if a Parakeet copy is already on the Mac.
• Fixed: macOS could close Heyra as idle when no window was open.
• The engine can run Whisper, Canary, Moonshine, Qwen3-ASR, Cohere Transcribe and NeMo
  CTC models from a folder of your own.

Speech model mirror: Parakeet TDT 0.6B v2 (int8)

Unmodified mirror of the sherpa-onnx int8 conversion of NVIDIA Parakeet TDT 0.6B v2. SHA-256 157c157bc51155e03e37d2466522a3a737dd9c72bb25f36eb18912964161e1ad. Model by NVIDIA, CC BY 4.0.

Heyra 0.5.4

• Labels in sentence case, in the interface's own face: no more spaced-out capitals
  in the sidebar, settings, cards, buttons or the copy card.

Heyra 0.5.3

• Home's status is a window into space: the orb itself, drifting, alive when you talk,
  with your mic and model as pills that open menus.
• Settings in iOS-style groups, with the microphone and voice model as menus.
  History shows every take in full.
• Theme at the top right: Dark, Light or Match the Mac.
• Tooltips spring in. Rows highlight with a rounded pill.
• An × on the words above the dot turns them off.
• After the download, a model says SETTING UP while it gets ready.
• Fixed: clicking the dot's panels could keep the keyboard, so takes didn't paste.
• Fixed: a thin white line above the words box on dark screens.

Heyra 0.5.2

• Heyra keeps itself up to date: once a day it checks for a new version and, if you
  installed it with Homebrew, upgrades and restarts into it by itself. No terminal.
• Home in pictures: today's words over a fortnight of bars, your streak as the last
  seven days, your pace against typing.
• Tooltips that show at once: beside the sidebar icons, above bars and days.

Heyra 0.5.1

• A new look: a top bar, a sidebar that folds down to icons, more room and fewer words.
  New type, with titles in the accent colour.
• Light and dark: Settings → Appearance (Dark, Light or Match Mac).
• Home shows today's words, time saved against typing, your streak and your pace.
• Hover the dot: Dictate, with the fn key. The pointer turns to a hand over it.
• After a brew upgrade, Heyra restarts into the new version by itself, between takes.
• With nowhere to type (Finder in front, no text field), your words wait above the dot
  with a Copy button.
• The words above the dot sit in two clean lines.
• A bigger window. The version is at the foot of Settings.

Heyra 0.5.0

• Heyra writes while you talk, so your text is ready about half a second after you let go,
  however long you spoke. A 90-second take used to take 7 seconds.
• Your words appear above the dot as you talk, and the finished text stays a moment once
  it's pasted. Turn it off in Settings → Show the words while you talk.
• A bigger, roomier window. Settings starts with the voice model and says less.
• Smoother scrolling, and less work while Heyra sits idle.
• Updating from before 0.4.0: Heyra offers the new voice models once.
• Fixed: a new install could switch to Parakeet on its second launch.
• The logo is an h now, and the resting dot has a fine gold ring.

Heyra 0.4.0

• Choose your voice model on the welcome screen or in Settings: Bark 3.0 Clean (English,
  without the "um"s), the default on a Mac set to English; Fluid Speech Mini (English,
  every word); or Parakeet v3 (every word, 25 languages). Or point Heyra at a folder
  with a sherpa-onnx model of your own.
• Switching downloads the new model while the current one keeps working. A dropped
  download resumes where it stopped and retries by itself.

Heyra 0.3.11

• Fixed: some takes came back empty, or missed whole sentences. The speech engine is now
  sherpa-onnx 1.13.8.
• Fixed: long hands-free takes lost words. They now go in 15-second pieces.

Model: Fluid Speech Mini

Fluid Speech Mini by FluidVoice (CC BY 4.0), converted for sherpa-onnx.