Embodiments of the present invention provide a method and device for transmitting voice in a wireless system. The method includes: identifying, by a transmitter, each original voice encoding packet needed to be sent out with a number indicating playback order, and performing channel encoding on each identified original voice encoding packet to construct a voice session packet; establishing a voice session or voice data mixed session between the transmitter and a receiver; allocating a channel dynamically for the voice session or the voice data mixed session; sending, by the transmitter, newly-arrived voice session packets, delayed voice session packets, voice session packets needed to be re-transmitted, data session packets and control command packets according to pre-configured priority; receiving and detecting, by the receiver, the voice session packets, sending an NACK packet comprising number of a lost voice session packet to the transmitter to inform the transmitter to re-transmit the voice session packet, if it is confirmed that the voice session packet is lost; and putting voice session packets properly received into a jitter buffer controller at the receiver if the receiver is a terminal. In embodiments of the present invention, spectral efficiency and reliability of real-time voice services in a wireless multi-service transmission system may be improved while satisfying the Quality of Service (QoS) requirements of real-time services, such as voice service.