摘要 |
PURPOSE: To conduct repeated readings without requiring the device to reset a decoder at a loop top during the repeated readings by setting predicting filter coefficients so that every time the results match with each other at the leading portion of the repeated readings during a decoding. CONSTITUTION: When a musical sound signal generating section 5 transmits compressed waveforms read from a waveform memory 6 to a decoder 7, a quantization width control block controls the quantized width in the decoder 7. At that time, a CPU 10 reads the prediction filter coefficient values related to the read-out musical sound waveforms of the section 5 from a prediction filter coefficient memory section 21 and sets them as the coefficients of each multiplier of the decoder 7. Thus, the data stored in a delayed register are read, prescribed computations are executed and desired predicted values are obtained. Thus, no device is required to reset at the loop top because filter coefficients having a high converging speed and a high stability are used and the contents of the delayed registers are matched to each other at the loop top. |