The Wav2Vec2FeatureExtractor currently only accepts input as a list of 1D NumPy arrays due to hardcoded padding logic in its base class SequenceFeatureExtractor. This restricts its use to raw waveform ...

The input should be a 2D NumPy array or 3D NumPy volume, of datatype float32. The values should range from 0 - 4096. NOTE: You can pass 2D inputs to 3D networks and 3D inputs to 2D networks, this ...