Recognises Voice Commands recorded on System Microphone
Dataset [link] of 120 files each, for 5 Voice Commands [back,forward,left,right,stop] was recorded by Lakshit
Each of those 120 files were augmented to 2002 files using augmentor.py , then put into full_dataset folder
Version 1 (v1) :
All audio files converted to frequency domain from time domain using MFCC coefficients, then trained different Random Forest Classifiers and Simple MLPs with varying parameters.
Version 2 (v2) :
Recorded Environment bias, trained on (recording,environment_bias) instead of just on recording. Got higher accuracy for same models as above.
Use record.py to test with laptop microphone. It first averages current environment bias, then lets you try voice commands.
Accuracy :
