Improvement And Realization Of Low Rate Speech Codec Based On Fixed-point Algorithm

With the development of communication technology, the speech transmit technology based on Internet have already enter in a real time era. Different platform have different requirement for the speech transmission. Speech coding is the way to meet these requirements.LPC10-e is a speech CODEC in the rate of 2.4k bps, which realizes relatively good quality of voice with a lower algorithm complexity and thus can be fit to low band network. This CODEC have a wide implication area .In order to realize it to a MCU which has limited resources, we must solve the fixed-point algorithm and related technical problems. The key techniques in fixed-point algorithm on MCU and the improvement of LPC-10e CODEC are discussed in this paper. When doing fixed-point transplant, we use Q16 calibration, log and exp table and small table for division algorithm after a large amount of test to improve the procession speed while not having a lot impact on the accuracy of the data. We also analyze the mental like sounds in fixed-point LPC-10e algorithm and make effective improvement. When doing the final test, we introduce PESQ to compare the speech quality after decoding in varies condition. The experimental results show that the quality of voice after decoded by fixed-point algorithm has no major difference compared with that decoded by floating-point algorithm.