Learning long-term filter banks for audio source separation and audio scene classification.