An encoding and decoding apparatus quantizes LSP (Line Spectrum Pair) parameters, which are characteristics parameters of spectrum information included in a voice signal, with high accuracy and stability. This encoding and decoding apparatus comprises a first quantizer in which quantization is performed independently in the unit of one frame, and a second quantizer which uses a correlation between adjacent frames. An error comparator compares quantization errors produced by the first quantizer and quantization errors produced by the second quantizer to select the one quantizer whose quantization errors are less than that of the other quantizer. The first quantizer produces a highly accurate quantization with stability regardless a condition of an input voice signal, while the second quantizer produces a highly accurate quantization when input voice signal stays quasi-stationary. By switching these two quantizers, this apparatus can offer stable and highly accurate quantization regardless of the condition of the input voice signal.