Towards Efficient and Scalable Speech Compression Schemes for Robust Speech Recognition Applications