About Voice AI

Voice AI is a solo project. I'm Stanislav Shupilkin — I build tools that live inside Telegram.

It started with a simple annoyance: voice messages you can't listen to right now. So I built a bot that turns them into text in seconds — and back: text into natural speech.

Before locking in the speech-to-text stack, I benchmarked 11 STT models (Whisper, Parakeet, ElevenLabs Scribe and more) on real voice memos, comparing accuracy, latency and privacy. I write up what I test on the blog.

Privacy is a design choice: transcripts aren't stored by default — the bot keeps only what it needs to count credits.

Feedback or a feature request? Message the bot — it reaches me.

More of what I build: smolevich.com.