Voice Recognition Train Model(KOI v1.12.0)

From KOI v1.12.0 onwards, saving and loading a voice recognition model is supported.

Install an SD Card

The trained model will be stored on an SD card, without an SD card we won’t be able to save this model.

Training a Voice Recognition Model

Loading Extensions

Search "KOI" in the search bar.

Or copy and paste the following link to the search bar.

https://github.com/KittenBot/pxt-koi

Blocks for Voice Recognition:

Sample Program:

Program Flow

1: Download the program to the Micro:bit.

2: Press A to record the first command. KOI’s screen will display green upon a successful record.

3: Press B to record the first command. KOI’s screen will display green upon a successful record.

AI Camera has the capacity for 10 voice samples.

4: Press A and B at the same time to save the model.

The saved model must be in a .json format, or else it will not work.

Sample Code

1. Voice Recognition Model Training(Firmware1.9.5_Extension0.5.7)

Extension Version and Updates

There may be updates to extensions periodically, please refer to the following link to update/downgrade your extension.

Makecode Extension Update

FAQ

1: There is no reaction after pressing the buttons on the Micro:bit.

· A: This is because KOI has a longer boot time than Micro:bit. When the power is turned on, Micro:bit has already ran the code for KOI initialization before KOI is ready.

· Solution: Reset your Micro:bit after KOI has been turned on. (The trick is to let KOI power on completely before initialization.)

2: Does KOI work with 3V input?

· A: No, KOI only works with 5V.

3: KOI displays red everytime I try to record my command.

· A: The noise level hasn’t been calibrated, it is essential to run the block before voice recognition.