Repolode
Submit
Infatoshi

Infatoshi/phonon

Open-source macOS voice typing that transcribes speech and cleans it up locally on Apple silicon, then types into the active app.

Open-source alternative to:Wispr Flow

View on GitHubWebsite
Stars
319
Forks
30
Open issues
7
Last commit
2 days ago

What is it?

Phonon is a voice typing app for macOS on Apple silicon (macOS 14 or later). You hold a key, speak, and release it to get text pasted into the frontmost app. Speech is transcribed by Parakeet TDT 0.6B v2, and a 1.7B correction model then rewrites the transcript into what you meant to type: misheard terms are fixed, filler words and false starts are removed, and only the final version of a self-correction is kept.

The correction style follows the app you dictate into, with casual, formal, prompt, and minimal modes. Both models run locally after a one-time download of about 4.9 GB on first launch. Apart from a daily check for new releases, which can be turned off, the app works offline. Phonon also includes a personal dictionary, a local history with optional retention limits, and a CLI for managing terms and stored recordings. A Windows build is in beta.

Key features

  • 01Transcribes speech locally with Parakeet TDT 0.6B v2 on Apple silicon.
  • 02Corrects transcripts to remove filler words, false starts, and misheard terms.
  • 03Picks a writing mode based on the app you dictate into.
  • 04Streams recognition while you speak, so text appears quickly at key release.
  • 05Includes a personal dictionary and a command-line tool for managing it.
  • 06Offers a Windows beta build.

Use cases

  • Dictating messages, emails, and documents on a Mac without sending audio to a cloud service.
  • Writing prompts for AI assistants and commands in a terminal by voice.
  • Adding technical terms and names to a dictionary so they are transcribed correctly.

Quick start

  1. 1

    Install the app with Homebrew

    brew install --cask infatoshi/tap/phonon
  2. 2

    Launch Phonon (first launch downloads about 4.9 GB of models)

    phonon

Things to know

  • macOS on Apple silicon (macOS 14 or later) is the shipped platform; Windows is beta.
  • First launch downloads about 4.9 GB of models.
  • The refiner model uses the LFM Open License, which limits commercial use to organizations under $10M annual revenue.
  • The Windows build is not code signed and triggers a SmartScreen warning.

In collections

Run LLMs locallyOpen-source Mac appsAI video & audio tools

Topics

#speech-to-text#voice-typing#local-ai#macos#mlx#dictation#apple-silicon

FAQ

What is phonon?
Open-source macOS voice typing that transcribes speech and cleans it up locally on Apple silicon, then types into the active app. Phonon is a voice typing app for macOS on Apple silicon (macOS 14 or later). You hold a key, speak, and release it to get text pasted into the frontmost app. Speech is transcribed by Parakeet TDT 0.6B v2, and a 1.7B correction model then rewrites the transcript into what you meant to type: misheard terms are fixed, filler words and false starts are removed, and only the final version of a self-correction is kept.
Is phonon free to use? What license does it have?
phonon is open source under the GPL-3.0 license. Check the license terms before using it in commercial products.
Is phonon still maintained?
Actively maintained. The last commit to phonon was on Nov 5, 2026.
What can I use phonon for?
Dictating messages, emails, and documents on a Mac without sending audio to a cloud service. Writing prompts for AI assistants and commands in a terminal by voice. Adding technical terms and names to a dictionary so they are transcribed correctly.
How do I install phonon?
Install the app with Homebrew: brew install --cask infatoshi/tap/phonon Launch Phonon (first launch downloads about 4.9 GB of models): phonon
Where can I get phonon?
The source code and installation instructions are on GitHub: https://github.com/Infatoshi/phonon

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects