A complex Ferrari LPC to LSF implementation
Jatin V. Domadiya, Amaresh P. Kandagal, N. Ramesh · 2014
The line spectral frequency (LSF) is used in speech compression because of their quantization property, which is computed using linear prediction conversion (LPC). Paper describes the faster conversion of LPC to LSF method and its implementation. Proposed method is based on complex Ferrari equation and implemented in MatLab, C floating and fixed point arithmetic. This algorithm does not require any further iterations to calculate the roots as in complex free Ferrari based solutions, or intra frame ordering property or a distribution probability of the interval of roots of LSF polynomials. Proposed algorithm has been compared to GSM AMR-NB speech codec method, Cardano's cubic solution based method and complex free Ferrari solution based methods up to 20thorder. The algorithm uses decimation in degree (DID). There is a reduction of 80% in iterations compared to traditional grid method used in AMR-NB for a polynomial of 10thorder.