GitHub - singhroshan1999/voice-classifier: Voice-Classifier is Artificial Neural Network based audio classifier, goal of which is to classify different voice (say human or animal) into their respective labels (like voice of different persons)

Voice classifier

Voice classifier is Artificial Neural Network based classifier who's goal is to classify different voices or sound against provided labeled training dataset which consist of doctored RAW audio file.
For training all RAW Audio files must be of same size preferably atleast 30 second duration
RAW audio training data must not contain any silence
This model is trained and tested in 8-bit unsigned PCM RAW audio format.

Audio Format

RAW audio format is sampled at 44100Hz which means amplitude of sound wave is taken 44100 times every second.
Amplitude is divided into 256 parts (in 8-bit PCM format) and stored in RAW audio file

NOTE: 44100Hz is chosen because if nyquist rate f_c > 2f_m where f_m = 20,000Hz (max human hearable frequency)

Every voice (say human voice) have distinct spectrum of frequency of harmonices (Hz) and loudness (dB)
It is observed that around 1024 sample is optimum for distinguishing several voices.
But because of sampling rate 44100Hz sample size should be multiple of 44100 because if we multiply sample rate with time t. we get total sample which is divided to get total training set size which should be integer
(44100t)/n = total_tarining_set
Nearest integer is 882

Name		Name	Last commit message	Last commit date
Latest commit History 6 Commits
.ipynb_checkpoints		.ipynb_checkpoints
__pycache__		__pycache__
example		example
img		img
VoiceClassifier.ipynb		VoiceClassifier.ipynb
VoiceClassifier.py		VoiceClassifier.py
readme.md		readme.md