audio2vec. An audio embeddings model that models sentiment into a 256 dimensional space through spectrograms extracted from raw waveforms.