Parakeet got faster from a small CPU/GPU coordination change. On LibriSpeech: 76.7x RTF → 86.5x RTF, same WER. About 10-12% less wall time, no model change. Coming in the next Wasper release. #nvidia#parakeet#transcription#macos#speechtotext
Your Mac can now transcribe speech 72× faster than real-time - fully offline.
Wasper 1.5.0 introduces Parakeet, a new on-device voice engine. Runs entirely on your Mac's GPU. No cloud, nothing leaves your device.
GPU is faster than CPU. It's obvious, but for some reason, nobody implements Parakeet TDT on this yet.
Wasper will receive support for Nvidia's model on a GPU engine (Metal) 🔜
#NVIDIACorporation#parakeettdt
For months, I've stopped typing. I dictate everything — Slack, emails, even code — and @Wasper_app turns it into text instantly.
Built it because I wanted Mac dictation that's accurate AND doesn't ship my audio to the cloud.
Wasper is a privacy-first voice dictation for Mac.
Speak in any app — Slack, Gmail, VS Code, Notes — and it types. Dictation, AI rewriting, and live translation all run locally. No cloud, no subscription required.
Built for people who think out loud.