Enhanced Speech Segregation with Low-Latency DNN processing