Skip to content

Repository files navigation

Golosok: Local Speech Recognition for macOS

Read this in another language: Русский

Awesome Swift macOS Apps Latest Release License

Native voice dictation powered by Sber GigaAM v3. Inference on Apple Silicon.
Instant text input via ⌥ + Space and offline media file transcription.

Golosok Demo 1

Golosok Demo 2

🌟 Recognition


Features & Architecture

  • Full Autonomy (Zero Cloud). Inference runs 100% locally. Audio stream never leaves your Mac. Total privacy.
  • Direct Input. Press ⌥ + Space — dictate — text drops instantly under your active cursor. Works in any app: VS Code, Notion, Telegram, Xcode, Obsidian, Slack, and more.
  • Hardware Acceleration. The model is optimized specifically for Apple Silicon (M-series). Execution is offloaded to CoreML / Apple Neural Engine (ANE) without CPU overheating.
  • Context & Slang Awareness. Excellent understanding of Russian, English, IT slang, code-switching ("Runglish"), and automatic punctuation placement.
  • Batch File Processing. Offline transcription for audio/video files of any duration (MP3, WAV, M4A, MP4, WebM).

Technical Specifications

Component Specification
OS macOS 13.0+ (Ventura, Sonoma, Sequoia)
Hardware Apple Silicon (M1 / M2 / M3 / M4 / M5)
Model Engine GigaAM v3 RNN-T (GGUF Q8_0) / Whisper Large v3
RAM (Idle) ~70 MB
RAM (Active Inference) ~100 MB
Distribution Size ~270 MB

Installation

  1. Download the latest Golosok.dmg from the Releases section.
  2. Drag Golosok.app into your /Applications folder.

Bypassing macOS Gatekeeper:
This project is distributed as open-source software without an annual paid Apple Developer ID certificate. To prevent macOS from blocking the first launch, remove the quarantine attribute via Terminal:

xattr -cr /Applications/Golosok.app

(GUI Alternative: Control + Right-click on the app icon -> Open -> Confirm).

System Permissions

On the first launch, the app will request two macOS permissions:

  1. Microphone — for capturing voice input.
  2. Accessibility — for automated text insertion (Cmd + V) directly under your active cursor.

(If you are updating from an older version and the Accessibility toggle seems frozen — remove Golosok from the system list using the - button and re-add it).

Controls & Shortcuts

Golosok runs silently in the background. Control is bound to global hotkeys:

  • ⌥ + Space (Option + Space) — Start / Stop Dictation
    • 1st Press: Activates recording (white island UI with green equalizer).
    • 2nd Press: Stops recording. Triggers inference (~0.1s) and pastes text into your focused input field.
  • Push-to-Talk / Hold Mode (Selectable in Settings)
    • Hold ⌥ + Space to record, release to instantly transcribe and paste.
  • ESC — Cancel
    • Instant hard-reset of the current recording without pasting or saving.

A menu bar icon (waveform) provides access to recent dictation history, one-click clipboard copying, and app settings.

Building from Source

If you want to clone the repository and build the Xcode project manually:

# 1. Clone the repository
git clone https://github.com/Berliner187/Golosok.git
cd Golosok

# 2. Download GigaAM model weights
curl -L -o Golosok/gigaam.gguf https://huggingface.co/memoravox/gigaam-v3-e2e-rnnt-gguf/resolve/main/gigaam-v3-e2e-rnnt-Q8_0.gguf

# 3. Open project in Xcode
open Golosok.xcodeproj

In Xcode, press ⌘ + R to build and run.

About

Free Wispr Flow & Superwhisper alternative. Native macOS voice dictation & speech-to-text. Powered by Sber GigaAM v3 & Whisper. Instant direct input under cursor via ⌥+Space, push-to-talk, offline transcription of any media files.

Topics

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Contributors

Languages