Multi-channel can be used to provide higher transmission ability to the bandwidth-intensive and delay-sensitive real-time streams. However, traditional channel capacity theories and coding schemes are seldom designed for the real-time streams with strict delay constraint, especially in multi-channel context. This paper considers a real-time stream system, where real-time messages with different importance should be transmitted through several packet erasure channels, and be decoded by the receiver within a fixed delay. Based on window erasure channels and i.i.d.(identically and independently distributed) erasure channels, we derive the Multi-channel Real-time Stream Transmission(MRST) capacity models for Symmetric Real-time(SR) streams and Asymmetric Real-time(AR) streams respectively. Moreover, for window erasures, a Maximum Equilibrium Intra-session Code(MEIC) is presented for SR and AR streams, and is shown able to asymptotically achieve the theoretical MRST capacity. For i.i.d. erasures, we propose an Adaptive Maximum Equilibrium Intra-session Code(AMEIC), and then prove AMEIC can closely approach the MRST transmission capacity. Finally, the performances of the proposed codes are verified by simulations. Multi-channel can be used to provide higher transmission ability to the bandwidth-intensive and delay-sensitive real-time streams. However, the traditional channel capacity theories and coding schemes are seldom designed for the real-time streams with strict delay constraint, especially in multi-channel context. This paper considers a real-time stream system, where real-time messages with different importance should be transmitted through several packet erasure channels, and be decoded by the receiver within a fixed delay. Based on window erasure channels and iid (identically and independently distributed) erasure channels, we derive the Multi-channel Real-time Stream Transmission (MRST) capacity models for Symmetric Real-time (SR) streams and Asymmetric Real- , a Maximum Equilibrium Intra-session Code (MEIC) is presented for SR and AR streams, and shown as able to asymptotically achieve the theoretical MRST capacity. For iid era sures, we propose an Adaptive Maximum Equilibrium Intra-session Code (AMEIC), and then prove AMEIC can approach approach the MRST transmission capacity. Finally, the performances of the proposed codes are verified by simulations.