Teachable Machine Local

Train a sound classifier with your microphone. Nothing is uploaded — everything runs on this machine.

1. Classes

Loading speech model…

Background Noise teaches the model what "nothing" sounds like — record at least 8 samples of your normal room sound. Each sample is 1 second.

2. Training

Record at least 8 samples per class.

Advanced

3. Preview

Train a model to start listening.

Exports the standard Teachable Machine audio format, loadable by @tensorflow-models/speech-commands.

Reopen a previously exported audio model to test it or stream to a micro:bit — no retraining needed.

4. micro:bit

Train a model first, then connect.

Sending: —

Background Noise never sends — silence means "nothing heard".