An implementation of the MatchboxNet: 1D Time-Channel Separable Convolutional Neural Network Architecture for Speech Commands Recognition architecture. No training logic, only the model (i.e., only a forward pass through the untrained model is possible at the moment).
from MatchboxNet import MatchboxNet
import torch
model = MatchboxNet(B=3, R=2, C=64, NUM_CLASSES=30)
sample_mfcc_features = torch.randn(1, 64, 465)
predictions = model(sample_mfcc_features)
print(predictions.shape)