SpeakoFlow Files

SpeakoFlow 2.2.0

A new speech engine with every model in Handy's catalog, other audio that gets quieter instead of muted while you dictate, Insights for the assistant and meetings with cards you can share, and mouse buttons as shortcuts, now out of Experimental on Windows and macOS.

On 2.0.0 or later? SpeakoFlow offers this update by itself, or you can check in Settings, About. On 1.4 or earlier? Download 2.2.0 below and install it over your current version. Your settings and history are kept.

Using Parakeet V2 or V3, Canary, Moonshine, SenseVoice, GigaAM or Cohere? These models run on the new engine now and need their new version. After updating, a pop-up offers Download new version, which gets the same model and switches to it when it's done. Nothing downloads until you choose.

New

• A new speech engine, and every model in Handy's catalog. The Models page opens on Recommended: the models you have, then the recommended ones you don't. Pick a language and it suggests models for that language. All models lists all 69, with filters for streaming, translation and language. Every download is checked against its checksum.
• Speech models run in their own process. If a graphics driver crashes, SpeakoFlow moves speech to the processor and finishes the dictation instead of closing (#82).
• Speech processing, in Settings, Advanced, Performance: Auto, a specific graphics card, or Processor. It replaces the old Whisper and ONNX acceleration settings and is no longer hidden behind Experimental.
• Split long recordings. A dictation longer than 40 seconds is transcribed in pieces cut at pauses, which keeps models that aren't built for long audio accurate.
• Lower other audio while you dictate. Settings, Audio, Other audio: Unchanged, Lower or Mute. Lower fades music, videos and calls down by an amount you choose, and back up when you stop.
• Insights covers the assistant and meetings. Share turns your numbers into cards you can copy, save, or post to X or LinkedIn. The cards are made on your computer and nothing is uploaded.
• Chinese script. Simplified, Traditional, or As transcribed, even with the language on Auto.
• Custom words now reach every model that accepts a vocabulary while it transcribes.

Changed

• Mouse buttons as shortcuts is out of Experimental on Windows and macOS. Turn it on in Settings, Shortcuts, Mouse. The middle, Back and Forward buttons work, and a button goes next to a shortcut's keys, so the keys keep working too. On Linux it stays under Experimental.
• Streaming models stream with the Minimal pill too, so the text is ready the moment you stop.
• Your language choice is kept when you switch models.

Fixed

• Moonshine Streaming cut dictations short or pasted the wrong words.
• A paste could insert what you had copied before in a busy browser tab or editor.
• Mute while recording could leave the computer muted after a very short dictation.
• Esc during "Transcribing…" now stops the transcription and offers Undo.
• Assistant answers pasted as raw Markdown in Word, Outlook, Gmail and Google Docs. They now paste with their formatting.
• Linux: memory grew with every dictation, and shortcuts stopped working after a keyboard reconnected or the computer woke from sleep (thanks to @ishangautam7).

Good to know

• Mute while recording is now Other audio, in Settings, Audio. If you had it on, it's set to Mute.
• Whisper .bin files you already downloaded keep working and are marked Older. Their newer versions are in the catalog.

Read the full [changelog](https://github.com/AbhishekBarali/SpeakoFlow/blob/main/CHANGELOG.md#220

Which file do I download

The other three files are not for you to download. Installed copies of SpeakoFlow fetch them by themselves to update: latest.json tells the app a new version exists, and SpeakoFlow_aarch64.app.tar.gz / SpeakoFlow_x64.app.tar.gz are the update packages for Apple silicon and Intel Macs. Windows: if SmartScreen warns you, choose More info, then Run anyway.
macOS: if it says the app "is damaged", run this once in Terminal (why?):

:::bash
xattr -dr com.apple.quarantine /Applications/SpeakoFlow.app