Skip to content

Repository containing the code to train a model to detect 2 words, silence, or unknown sounds and the code for deploying the model to an arducam Pico4ML embedded system

Notifications You must be signed in to change notification settings

AustenBrooks/wake-word-detection

Repository files navigation

wake-word-detection

Repository containing the code to train a model to detect 2 words (right and bamboozle), silence, or unknown sounds and the code for deploying the model to an arducam Pico4ML embedded system

micro_speech

micro_speech contains all the C++ code for deploying the model to the board, this is based off of the https://github.com/ArduCAM/pico-tflmicro.git, view the quick start guide at https://www.arducam.com/docs/pico/arducam-pico4mltinymldevkit/quick-pico-setup/#micro-speech for more instructions

bamboozle

folder containing custom recordings of bamboozle

bamboozle recorder

jupityr notebook for recording audio segments in the correct format to match the google dataset

micro_speech.uf2

generated file to be uploaded to the board

model training

jupityr notebook that uses the custom recordings and the google speech dataset (multiple variants available) to train and save a tensorflow model, also creates a tflite file which contains the hex values used in the micro_speech\micro_features\model.cpp file (this is how you upload a custom trained model)

pitch shifting

jupityr notebook to pitch shift audio files (nice when using small amount of custom data)

streaming audio

test script to run the model on a computer before deploying to a board

About

Repository containing the code to train a model to detect 2 words, silence, or unknown sounds and the code for deploying the model to an arducam Pico4ML embedded system

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published