Simulation And Research Of Adaptive Multi-Rate Wideband Speech Coding Algorithm

In order to meet the requirement of WCDMA and GSM in high quality speech services, 3gpp/ETSI proposed the adaptive multi-rate wideband (AMR-WB) speech codec. After that, the AMR-WB codec was selected by ITU-T in the standardization activity for wideband speech coding around 16 kbit/s. AMR-WB has the advantages of high speech quality, low average bit rate and strong adaptability. The adoption of AMR-WB by ITU-T is of significant importance for it is the speech code system adopted both by wireless and wire line service at first time, and will be used widely in wireless communication fields as well as wire line communication fields.This paper systemically analyzed the AMR-WB algorithm. Based on deep research the liner prediction and quantization, adaptive codebook search, fixed codebook search, high band process and the decoding principle, the AMR-WB algorithm was simulated. The standard speech files (16kHz, 16bit) of TIMIT standard English speech database were selected to accomplish coding and decoding experiments, then tested the synthesized speech in subject auditory quality and object PESQ value. The result showed above the model of 12.65kbit/s, the waveform of synthesized speech is the same as that of input speech, and it had hardly to differentiate with original input speech in auditory, the PESQ values were more than 4.0. In the model of 6.60kbit/s and 8.85kbit/s, there were a little distortion in terms of waveform and auditory, and the PESQ values were above 3.5.They all reached the communication quality standard, and the synthesized speech had the good naturalness and comfortable auditory property.In this paper, three kinds of wideband speech coding standard of G.722, G722.1 and G722.2 (AMR-WB) also were tested, and the speech quality of these algorithms was compared. The result presents the speech quality in the model of 23.85kbit/s and 23.05kbit/s in G.722.2 (AMR-WB) was equivalent with 64kbit/s in G.722, the speech quality in the model of 19.85kbit/s and 18.25kbit/s in G.722.2 (AMR-WB) was equivalent with 56kbit/s in G.722, and the model of 32kbit/s in G.722.1 was the same as the model of 12.65kbit/s in G.722.2(AMR-WB). Apparently, G.722.2 used the lower bit rate to obtain the higher speech quality, and there was obvious advantages compared with G.722 and G.722.1.A correction of perceptual weighted filter was used in the AMR-WB algorithm, this perceptual weighted filter was different from the traditional narrowband speech perceptual weighted filter, it can combine with pre-emphasis to reduce the spectrum inclination caused by the large dynamic range from low frequency to high frequency of wideband speech signal, the new perceptual weighted filter could reduce the inclination of formant. This paper analyzed and studied the two perceptual weighted filters, and compared their frequency characteristics by made full use of standard wideband speech. The result showed that this new perceptual weighted filter had better filtering characteristics, and the frequency response could be well corresponding with the spectral envelope of wideband speech, moreover the smoothness property of this filter is better and the change of dynamic range is small.