The model supports 25 European languages when running on a laptop and 8 languages on mobile devices. In performance tests using an iPhone 17 Pro Max's Neural Engine, the 8 language version transcribed a 10 minute recording in 1.65 seconds. Tether's QVAC division noted that as phones, glasses, and robots shift toward voice interfaces, processing audio locally via the parakeet engine is critical for privacy and functionality.
Wednesday, October 21, 2026
Story
Tether Ships 254M Parameter AI Model for Faster On Device Speech
AI7 days ago
The company behind the USDT stablecoin has introduced a new speech to text tool that runs entirely on local hardware to protect user privacy. Known as QVAC TranscribePsy, the 254 million parameter model allows transcription on smartphones and laptops without an internet connection and eliminates the need to send audio data to cloud servers. The issuer of USDT stated that the model outperformed competitors in speed tests conducted across six different smartphones.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.