Krisp · depuis Juni 2019 (erste Produktveröffentlichung als Noise-Cancellation-App); seitdem kontinuierlich um Transkriptions- und Voic · 3× · vu le 30 juin 2026
Krisp is a Voice AI platform from Krisp Technologies, founded in Armenia in 2017 (now headquartered in Berkeley, USA), originally launched as a pure noise-cancellation tool. Today the product also includes an AI meeting assistant with transcription, summaries, and real-time accent conversion across desktop, mobile, and browser platforms. The software runs as a virtual audio device between microphone/speaker and communication apps like Zoom, Teams, or Google Meet, processing audio largely on-device. Pricing ranges from a free tier with usage limits to Business and Enterprise plans for organizations and call centers.
Fonctionnalités
| Real-Time Streaming | Yes, real-time noise cancellation, accent conversion, voice translation and live transcription during calls |
| Latency | Approx. 200ms latency for accent conversion, under 20ms for basic noise cancellation |
| Latency (ms) | ~20 ms minimum latency for the noise cancellation/voice isolation model (VIVA SDK, CPU-based, real-time audio stream) |
| Model Size (Parameter Count) | Turn-taking model: 6 million parameters (audio-only); generally, "very small, CPU-based AI models" are used (no published parameter count for the STT model itself) |
| Platform | Desktop (Windows, Mac, Linux), Web/Browser, iOS and Android; integrates with Zoom, Google Meet, Microsoft Teams, Slack and 800+ other apps |
| Price | Free: 60 min/day noise cancellation, unlimited transcription; Pro: $8/month (annual) or $12-16/month; Business: $15/user/month (annual); Enterprise: custom pricing |
| Release Date | First released June 2019 as an AI noise-cancellation app; company founded in 2017 |
| Languages | Meeting transcription supports 16-17 languages (incl. English, Spanish, French, German, Russian, Italian, Dutch, Polish, Portuguese, Hindi, Korean); Voice Translation supports 80+ languages |
| Voice Cloning | No voice-cloning feature; instead offers Deepfake Detection to flag synthetic/cloned voices in the contact center product |