Apparatus and method for echo control in parameter domain

The method according to 10, wherein detecting whether the near-end input signal frame is an echo frame according to the parameter domain characteristic parameters of the far-end output signal and the near-end input signal comprises:

• extracting the parameter domain characteristic parameters of the current sub-frames of the far-end output signal and the near-end input signal, respectively, so as to construct far-end vectors and near-end vectors, respectively, wherein the parameter domain characteristic parameters further comprise pitch period and energy;
• performing a cross-correlation operation on the far-end vectors and the near-end vectors to obtain groups of cross-correlation coefficients, and recording the timing corresponding to each group of cross-correlation coefficients;
• judging whether the current sub-frame is an echo sub-frame by detecting the similarity between far-end pitch period and near-end pitch period at the timing corresponding to the maximum cross-correlation coefficient, or by detecting whether the difference between far-end energy and near-end energy is below a predetermined threshold;
• and
• judging whether the near-end input signal frame is an echo frame according to the echo detection result of each sub-frame in the near-end input signal frame.

The method according to 10, wherein re-quantizing the LSF prediction error in the near-end input signal frame comprises:

• obtaining an LSF prediction value of the current near-end input signal frame by multiplying the LSF prediction error of a previous near-end input signal frame with an LSF prediction error coefficient and adding the multiplication result with an LSF average value, wherein the LSF prediction error coefficient and the LSF average value are both constants and determined according to the encoding mode of the near-end input signal frame;
• obtaining an LSF prediction error of the current near-end input signal frame prior to quantization by subtracting the LSF prediction value from the LSF of the near-end input signal frame;
• and
• quantizing the LSF prediction error.