An instantaneous amplitude model based speech coder

In this paper, a new algorithm for speech coding is proposed. This algorithm is based a revised sinusoidal model, in which each component is represented with two instantaneous amplitudes and a frequency. This model avoids the difficulty in estimating the highly nonlinear phases and allows one to optimize the amplitudes once the frequencies are estimated. Simulations indicate that the proposed model can represent speech signals very well. Furthermore, based on this model, a speech coder is developed, in which all the estimated magnitudes are approximated with a four-parameter model. With the four parameters, the obtained phases and frequencies encoded using the simplest linear (scalar) quantization, a 16.64 kb/s speech coder yielding high quality synthetic speech signals.