Bi-Sep: A Multi-Resolution Cross-Domain Monaural Speech Separation Framework