The invention discloses a voice signal processing method and device, and the method comprises the following steps: step 1, receiving a decoded far-end signal, and collecting a near-end signal while playing the decoded far-end signal; step 2, estimating and compensating the delay, and performing data alignment operation on the far-end signal and the near-end signal; step 3, sending the aligned far-end and near-end signals into an adaptive echo suppression unit with feedback, and suppressing echoes in the near-end signals; step 4, inhibiting residual echoes and howling in the near-end signal; step 5, suppressing the noise in the near-end signal; and step 6, transmitting the near-end signal through the network. Through the method and the device, the echo, the residual echo, the howling and the noise in the full-duplex call can be suppressed. The problems of echo, howling, noise and the like in the voice talkback can be solved with relatively low complexity.