TranscribePsy: on-device speech recognition, including Portuguese
On October 14, 2026, Tether AI Research released TranscribePsy, a 254M-parameter speech recognition model that runs on a phone or laptop without the audio leaving the device.
According to Tether, the model covers 25 European languages on a laptop and 8 on a phone, including Portuguese, and makes 6.58 word errors per 100 on Multilingual LibriSpeech, against 11.51 for Whisper-small. The weights are Apache 2.0, but Tether says they were released for research and education, and live streaming is not available yet.
For consultations, interviews and meetings with sensitive data, transcribing on the device avoids sending the recording to a third-party server. Before using it, test with audio in your accent and vocabulary and confirm the terms for commercial use.