NEW: Tether has launched a 254-million-parameter AI speech recognition model that transcribes audio entirely on-device without sending data to cloud servers. The USDT issuer said its QVAC TranscribePsy model outperformed rivals in speed tests across six smartphones.