GigaAM or Vosk: which speech recognition to pick

Both models come from Russia, handle Russian, run without a GPU and allow commercial use: GigaAM under MIT, Vosk under Apache 2.0. Sber's GigaAM covers more conversation analytics: emotion recognition, a v3 version with punctuation, and a multilingual version adding Kazakh, Kyrgyz and Uzbek, last updated 2026-07. Vosk is stronger where offline work on weak hardware matters: models from 45 MB run on a Raspberry Pi or a phone, there are streaming models for live audio, and Vosk TTS provides simple Russian speech synthesis. If you use Vosk TTS for voiceover, remember that a voice may only be cloned with the consent of the person it belongs to. The choice comes down to conversation analytics versus on-device autonomy.