In downlink of code-division multiple-access (CDMA) networks, the maximal number of real-time calls can be increased by smoothly increasing the end-to-end distortions. In this paper, a cross-layer optimization system is developed to control each user's distortion by adapting source coding rates, channel coding rates, and transmit powers. In addition, the channel-induced distortion is controlled to be only a small proportion of the total end-to-end distortion, such that the subjective quality of the received signal is high. The formulated problem is to reduce the overall end-to-end distortion in downlink single-cell systems, under the constraints of users' maximal acceptable distortions and maximal total transmit power from the base station. To solve this problem, a near-optimal algorithm is constructed to allocate resources. A performance upper bound is developed and compared with the performance of the proposed algorithm. A dynamic system considering speech activities and different offered loads is also analyzed. From the simulation results, the proposed algorithm significantly reduces distortion and the necessary maximal transmit power when the number of users is large, compared with the traditional voice over CDMA schemes.