An important part of the Voice Recognition is the MFCC processing which is based on the samples on my voice, therefore, the Convolutional Neural Network has been trained to work only with my voice, which is the wanted result, however, there are still chances of other speakers to be recognised.
I decided to perform some tests as it would be important to check the actual accuracy of the model with unknown speakers. After employing the help of 2 diferent speakers of different ages, mother tongue, pitch and tone, I got the following results:
In the figure can be seen how for the first speaker out of 280 voice commands tested, 51.66% were recognised. Meanwhile, for the second speaker out of 35 voice commands, 56.77% were recognised.
If the device would be sold to consumers, the voice model would require voice samples from multiple speakers of different backgrounds, mother tongue, pitch, tone and accents.