| 摘要 |
A method of quantizing the line spectral frequencies (LSF) when calculating the parameters of an analysis filter included in a LPC-speech coder. The line spectral frequencies form an alternative to the filter parameters with unambiguous correspondence. It is earlier known to construct sum polynomials (P) and difference polynomials (Q) from the direct form coefficients of filters and thereafter to calculate the roots of the polynomials which correspond to the line spectral frequencies. This calculation is carried out by means of iteration. According to the inventive method, the roots of the sum and difference polynominals are not calculated, and the polynominals are instead investigated with the aid of pro-selected test frequencies (ftp1' ftqf1' ftp2' ftq2'...) which are speech typical. The polarity of the two polynominals (P, Q) is investigated for each of these test frequencies. When a polarity change occurs between two test frequencies, there is chosen one of these frequencies, for instance the higher frequency, which gives a given root (p1 and q1 respectively) of the sum polynominal (P) for the difference polynominal (Q), and therewith a line spectral frequency (LSF1 and LSF2 respectively). This method obviates the need for complicated calculations and improves the use of the properties of the polynominals. <IMAGE> |
