Tether AI Research Releases 254 Million Parameter Edge Speech Recognition Model TranscribePsy

According to Techub News, Tether AI Research has released the open-source edge-side speech recognition model TranscribePsy. This model has 254 million parameters and can run offline on devices such as smartphones, smart glasses, and humanoid robots without needing to upload audio to the cloud for processing, aiming to protect user privacy. TranscribePsy leads in multiple benchmark tests. It takes only 1.65 seconds to transcribe 10 minutes of audio on the iPhone 17 Pro Max, which is 2.6 times faster than similar models. The model supports 25 European languages and has been optimized for the iPhone Neural Engine.