Research And System Implementation Of Speech Coding And Decoding Technology In Narrowband Low-Reliability Channels

Currently,in specific military and civilian applications,narrowband and low-rate communication methods such as shortwave are still indispensable means of transmitting speech in their channels.To achieve this,the development of high-quality,low-rate speech coding technology is continuously required,especially for speech coding technology operating at 2.4kbps and lower rates.The 10 th Linear Prediction Coding(LPC)algorithm,as a milestone algorithm for speech compression coding,is based on the principle of human vocal characteristics and serves as the foundation of existing speech compression coding algorithms.However,due to technical limitations,speech synthesized by the LPC 10 algorithm contains buzzing and other noises,resulting in poor auditory perceptual quality.With the continuous improvement of speech quality requirements and the rapid development of communication technology,speech coding is also moving towards higher quality and lower rates.This dissertation is based on the LPC 10 algorithm and focuses on researching higher quality and lower rate speech codec technology.A mixed excitation algorithm based on Linear Spectral Frequency(LSF)parameters is proposed and implemented.The main research work is as follows:(1)The algorithm framework was reconstructed.Based on the LPC 10 framework,the encoding end introduces new transmission parameters for Fourier spectral amplitude and sub-band voiced/unvoiced intensity.At the decoding end,the excitation source generation method is changed from binary excitation to mixed excitation.Additionally,adaptive spectral enhancement technology is added to improve the synthesized signal waveform,making it more closely matched to the original signal waveform.(2)The encoding method and feature parameters were optimized.At the encoding end,the processing and transmission of linear predictive coefficients are optimized by replacing reflection coefficients in the LPC 10 algorithm with line spectral pairs frequency parameters,which are then subjected to four-stage vector quantization to reduce the coding rate.At the decoding end,parameter recovery is first performed by reference to the codebook,and linear predictive coefficients are obtained by solving the calculation,minimizing the loss between quantization and de-quantization.This ensures that the reconstructed linear predictive coefficients have a lower degree of distortion,which ultimately enhances the naturalness and intelligibility of the synthesized speech when applied to the synthesis filter.(3)The excitation source was improved.At the encoding end,parameters for sub-band voiced/unvoiced analysis are analyzed and transmitted for voiced frames.At the decoding end,the binary excitation in LPC 10 is improved to mixed excitation.The pulse sequence and noise sequence for generating the fundamental period are shaped by filtering,and the weighted sum of the pulse and noise sequences is calculated based on the sub-band voiced/unvoiced intensity to generate mixed excitation source.When applied to the synthesis filter,this method reduces the buzzing sound in the synthesized speech,improves continuity,and enhances the fullness and detail of the sound.(4)An embedded implementation of the algorithm was carried out.The improved algorithm is implemented in C and verified on an embedded board,achieving a complete process of speech input,encoding processing,bit transmission,decoding processing,speech synthesis,speech playback,and speech quality evaluation.The verification results show that the speech coding and decoding algorithm is stable and can meet the requirements of narrowband low-rate communication.This dissertation proposes an improved mixed excitation algorithm based on LSF parameters,with an encoding rate of 2.1kbps,which is approximately 12.5% lower than the LPC 10 algorithm’s2.4kbps.Using the Perceptual Evaluation of Speech Quality speech quality evaluation system to evaluate the speech before and after encoding,the results show that the speech quality score obtained by the proposed algorithm is 2.35,while the LPC algorithm’s score is 1.30.The proposed algorithm improves the LPC 10 algorithm by 1.05 points(out of 4.5 points)in terms of speech quality.