Sound Classification Exploration

Exploring sound classification methods

Requirements

Create a python virtual environment and install requirements

pip install -r requirements.txt 

Clone the repository ESC-50

Download Audioset embeddings models below and place them in audioset/ directory

• VGGish model checkpoint, in TensorFlow checkpoint format.
• Embedding PCA parameters, in NumPy compressed archive format.

Feature extraction

Edit src/feature_extract.py and set the following values

DATASET_DIR = "ESC_50/audio/"
DEST_ROOT_DIR = '../features/'

# one of ['spectral_feat', 'mel_stft', 'mel_stft_db', 'stft', mfcc_zcr', 'audioset_em']
FEATURE_TYPE = 'stft'
NUM_PARALLEL_PROCESS = 8

SAMPLE_RATE = 44100
HOP_SIZE = 512
WINDOW_SIZE = 1024
N_MELS = 128

Running python feature_extract.py will save the features for each audio file into a numpy array saved in .npy format in DEST_ROOT_DIR.

Sound classification methods

Explore sound classification methods in following notebooks

• src/spectral_features.ipynb : Spectral features (centroid, bandwidth, flatness, rolloff)
• src/mfcc.ipynb : Mel-frequency cepstral coefficients (MFCC)
• src/stft.ipynb : short-time Fourier transform (STFT)
• src/audioset_embeddings.ipynb: Audio embeddings as feautres