Golosok
Golosok is a free, open-source dictation app for macOS under MIT by Kozak Developer. It runs on Apple Silicon and transcribes Russian mixed with English technical words through Sber's GigaAM v3 model, with Whisper Large v3 as an option. Audio stays on the Mac and the README gives about 0.1 seconds latency on M-series chips. One caveat: Apple Silicon only and the main target is Russian; English-only users have more polished choices.
What does Golosok do?
Golosok describes itself as "Native voice dictation powered by Sber GigaAM v3. Inference on Apple Silicon." Press Option plus Space, speak, release, and the text is inserted at the cursor. A second mode batch-transcribes an audio or video file to a transcript.
The default speech model is GigaAM v3, Sber's Russian-first speech model, in RNN-T form, quantised to Q8_0 and run through a bundled llama.cpp build. Whisper Large v3 is available as an alternative engine. Both are supplied as downloadable weights, and both run on the Mac with no network call for transcription. The README names Russian, English, and code-switching between them (reading a Russian sentence with English technical words) as the target.
Key facts
- Licence: MIT, copyright 2026 Kozak Developer.
- Platform: macOS 13 Ventura or later on Apple Silicon, from M1 to M5. Intel Macs are not supported.
- Default speech model: GigaAM v3 RNN-T, from Sber, in GGUF Q8_0 form.
- Alternative speech model: Whisper Large v3.
- Latency: about 0.1 seconds from release to text, measured by the maintainer on Apple Silicon.
- Memory: about 70 MB at rest and about 100 MB during inference, from the README.
- Install size: about 270 MB on disk for the signed app.
- Languages: Russian, English, mixed Russian and English technical words, with automatic punctuation.
- Install: download the disk image from the Releases page, drag the app to Applications. First launch may need
xattr -rd com.apple.quarantinebecause the download is not notarised.
What does it replace, and where does it fall short?
Golosok replaces Superwhisper and Wispr Flow for Russian speakers on Apple Silicon who want local transcription with no subscription.
Where it falls short: Apple Silicon only. Intel Macs, Windows and Linux are not supported, and the README states macOS 13 as the floor. The default model is Russian first, so an English-only user has little reason to pick GigaAM over Whisper, and the Whisper Large v3 choice is already available in several English-first apps on this directory. The GigaAM model is Apache 2.0 by the model card; the README does not document how the Whisper weights are packaged. The latency and accuracy claims come from the maintainer and were not verified here.
How does Golosok run?
Download the DMG from the Releases page and drag the app into Applications. The first time a notarised build is not available, the README says to clear the quarantine bit with xattr -rd com.apple.quarantine /Applications/Golosok.app. Grant Accessibility and Microphone permissions.
Press Option plus Space to start recording. Release to stop and paste. The current model is shown in the menu bar, and a settings pane switches between GigaAM and Whisper, each of which is downloaded once on first use.
Who is Golosok for?
A Russian-speaking developer or writer on an Apple Silicon Mac who wants to dictate Russian with English technical terms mixed in, keep audio on the device and not pay a subscription. An English-only user on Apple Silicon has more polished choices in Handy and VoiceInk; a user on Intel, Windows or Linux should look at OpenDictate, Plume or local-stt.
What limits does the README state?
- Apple Silicon only, macOS 13 or later.
- The downloaded disk image is not notarised at the time it was read, so first launch needs the quarantine bit cleared.
- Accessibility and Microphone permissions are required.
- English-only users get Whisper; GigaAM is the Russian-first choice.
- Benchmark figures (about 0.1 seconds, about 70 and 100 MB of RAM) come from the maintainer on one machine family.
Questions people ask
Is Golosok open source?
Yes. The LICENSE file is MIT, an OSI-approved licence, copyright 2026 Kozak Developer. The GigaAM v3 model weights it downloads carry Sber's own licence, recorded on the model's page on Hugging Face. Whisper Large v3 is OpenAI's MIT-licensed model.
Does Golosok work offline?
Yes. Both GigaAM and Whisper run on the Mac through a bundled llama.cpp build after the one-time weights download, and the README says audio stays on the device. There is no cloud option and no account step.
Can Golosok be used commercially?
Yes for the code, which is MIT. The GigaAM v3 weights are Apache 2.0 by the model card, which allows commercial use. Whisper Large v3's weights are MIT. A commercial user should still read each model card before shipping a product that embeds the weights.
How does Golosok compare with Superwhisper?
Golosok is free, open source and focused on Russian with English code-switching. Superwhisper is a paid Mac app that supports many languages and ships cleanup prompts. Both run the model locally; the closest match on this list for English users is Handy with Whisper or Parakeet. The open-source alternatives to Superwhisper page sets both beside Voz, VoiceInk and others.
Why does Golosok use GigaAM instead of Whisper by default?
The README's own answer is that GigaAM v3 is trained on Russian and handles Russian mixed with English technical terms better than a generic Whisper model. Whisper Large v3 is included as an option for languages GigaAM does not cover. A user whose main language is not Russian can switch to Whisper in the settings.
Sources
- Golosok README and LICENSE: github.com/Berliner187/Golosok, read 2026-11-09.
More voice dictation apps
A free, offline push-to-talk dictation app for macOS, Windows and Linux that runs Whisper or NVIDIA Parakeet on your own machine and pastes the text at your cursor, under MIT.
A GPL-3.0 macOS dictation app that runs Nemotron, Parakeet, Cohere Transcribe, Apple Speech or Whisper on the device, with command and write modes, per-app prompts, and an optional closed local enhancement model.
A dictation and meeting-notes app under MIT that runs Whisper or Parakeet locally or cloud models with your own key, detects Zoom, Teams and FaceTime calls, and labels speakers on device.
A macOS dictation app under GPL-3.0 whose README names Superwhisper and Wispr Flow as the tools it replaces, with local models, per-app modes, a personal dictionary and an AI assistant mode.
A speech-to-text app inside the Epicenter monorepo under AGPL-3.0 that records, transcribes with a provider you choose, cloud or local, optionally polishes the text, and pastes it at the cursor.
An Apache-2.0 macOS dictation app that calls Gemini 3.5 Transcribe with your own key, writes audio to disk from the first millisecond for crash safety, and inserts polished text at your cursor without an account.
Added November 9, 2026. Every claim above comes from the project's README, LICENSE or model card, read on November 9, 2026, or from the GitHub API on the date shown in the panel. Found an error? Write to reveneau@licheo.com and it is fixed in the next weekly pass. Repository: github.com/Berliner187/Golosok.
Want Golosok running on your own servers?
Reveneau sets it up on your hardware, connects it to your product, and builds the software around it.