On-device meeting transcription means the speech-to-text runs entirely on your phone, with no audio sent to a server. mind² ECHO transcribes meetings locally with the Whisper speech model and tells speakers apart on the device, so a recording of a private conversation never has to travel to anyone’s cloud.
Most meeting tools upload your audio to their servers to transcribe it. On-device transcription does the opposite: the model that turns speech into text lives in the app and runs on your phone’s own processor. The audio stream is read, transcribed and discarded locally. Nothing about the recording, not the audio and not even the raw transcript, has to leave the device for the words to appear.
Transcription uses the Apache-licensed ACFT Whisper models from FUTO, whose mission is freeing technology from the control of the few. Speakers are then told apart by the open-source sherpa-onnx stack, a pyannote segmentation model paired with wespeaker speaker embeddings. All of it runs offline on the phone, in real time, so you watch the transcript build as people speak.
A meeting is one of the most sensitive things you can record: salaries, strategy, health, personal conflict. Keeping transcription on-device means that sensitive audio is never exposed to a third-party server, a subprocessor, or a data breach you would never hear about. When you ask the room to agree to a recording, you can honestly say the audio stays on your phone. Only the transcript text, never the audio, is later sent for AI analysis, and only with your explicit consent; the server does not keep it.
Because mind² ECHO records through your device’s microphone, it works around a table just as well as on a video call. There is no meeting bot to invite into the call and nothing for the other participants to install. One tap starts the recording; the transcript and the per-speaker analysis follow.
