LightfieldAI-native CRM that builds itself and does work for you
Bzz — Not your ordinary voice-to-text
Local-first dictation. Your voice never leaves your PC
11 followers
Local-first dictation. Your voice never leaves your PC
11 followers
Most dictation apps stream your audio to their cloud. Bzz does the opposite — transcription runs locally on your machine, so your voice never leaves your computer. Press F9, talk into any app, and text lands at your cursor. No word caps, no session limits. Optional AI polish — local model, your own cloud key, or none. No server processing your audio means ~$4–8/mo, not $12–15. Free until Labor Day. Windows + macOS. From Hiventiq — human-centered AI, rebuilt.
How well does the local transcription handle accents or background noise compared to something like Whisper running on the same hardware?
@halitbozan92903
Fair question, but it's actually the same thing under the hood — bzz runs faster-whisper (the CTranslate2 port of OpenAI's Whisper) locally, so accent and noise handling is identical to Whisper of the same model size. Whisper is genuinely strong on accents because it was trained on very diverse multilingual audio; background noise is its weaker spot, same as vanilla Whisper.
Where bzz gives you an edge is that you can step up the model size (base → small → medium → large-v3) if your hardware allows — larger models are noticeably more robust on heavy accents and noisy rooms, and it's one click in settings.
How does the local transcription actually perform on something like a mid-range laptop without a discrete GPU, and is there a noticeable lag when switching between the AI polish modes?
@glhaneb63 Thanks for the note.
Runs fine without a GPU — bzz detects there's no CUDA and automatically drops to the base Whisper model with int8 quantization on CPU, so a typical mid-range laptop transcribes a spoken sentence in a couple of seconds rather than real-time-instant. If you have any NVIDIA card it picks it up automatically and gets much snappier.
Switching polish modes has zero lag by design — nothing reloads. Each mode is just a different instruction applied to your next dictation, so the switch itself is instant. The only latency you'll ever feel is the AI pass itself (and "verbatim" mode skips that entirely, pure local, no round-trip).
pressed F9 in notion and it just dropped clean text at my cursor, no lag and nothing touched the cloud. love that i can plug in my own api key when i want a polish pass
@aslaaayakugaa
That's exactly the design goal — the transcription never leaves your machine, and the AI pass only happens when you invoke it, through your key, to the provider you choose. No middleman markup on tokens, no forced subscription to unlock "AI features" you're already paying a provider for.
It's also why a desktop app can be priced the way it is: cloud dictation tools have to pay for GPU servers, audio pipelines, storage, and compliance overhead for every minute you speak — and that all lands in your subscription whether you use it or not. Your laptop already has the compute; bzz just uses it. BYOK means the same thing for the AI side — you can even point it at a local model (Ollama works) and keep the polish pass fully offline too.