matlab-基于matlab的语音编码器实现人类语音信号合成-源码

clc; clear; close all; warning off; addpath(genpath(pwd)); %**************************************************************************** %更多关于matlab和fpga的搜索“fpga和matlab”的CSDN博客: %matlab/FPGA项目开发合作 %https://blog.csdn.net/ccsss22?type=blog %**************************************************************************** [A, Fs]= audioread('dat.wav'); A = A(:,1); frame_size = 0.030; % 30 ms frame_length = Fs*frame_size; percent_overlap = 0; % 50% overlap frame_overlap = frame_length*percent_overlap/100; frame_step = frame_length - frame_overlap; audio_length = length(A); no_of_frames = floor(abs(audio_length - frame_overlap)/(abs(frame_length - frame_overlap))); rest_samples = mod(abs(audio_length - frame_overlap) , abs(frame_length - frame_overlap)); % padding if remaining samples != 0 if rest_samples ~= 0 pad_signal_length = int16(frame_length - rest_samples); z = zeros(pad_signal_length,1); %A(end+1:end+pad_signal_length)= 0; A = vertcat(A, zeros(pad_signal_length,1)); no_of_frames = no_of_frames + 1; end % framing mat = zeros(no_of_frames, frame_length); i = 1; for r = 1:no_of_frames for c = 1:frame_length mat(r,c) = A(i); i = i+1; end i = i-frame_overlap; end % hamming window coefficients hamming = zeros(1,frame_length); for i=1:frame_length x = 2*pi*(i-1); x = x./(frame_length-1); hamming(1,i)=0.54-0.46.*cos(x); end % windowing for r = 1:no_of_frames mat(r,:)= mat(r,:).*hamming; end % computing ste and ZCR ZCR = zeros(no_of_frames,1); STE = zeros(no_of_frames,1); for r = 1:no_of_frames STE(r) = sum(mat(r,:).^2); ZCR(r) = mean(abs(diff(sign(mat(r,:))))); end STE_thresh = 0.456; ZCR_thresh = 0.5; OUTPUT_STE = zeros(no_of_frames,1); for i = 1:no_of_frames if STE(i) > STE_thresh OUTPUT_STE(i) = 1.0; else OUTPUT_STE(i) = 0.0; end end OUTPUT_ZCR = zeros(no_of_frames,1); for i = 1:no_of_frames if ZCR(i) > ZCR_thresh OUTPUT_ZCR(i) = 0.0; else OUTPUT_ZCR(i) = 1.0; end end % output array for plot o_arr1 = zeros(length(A),1); j = 1;k = 1; for i = 1:no_of_frames while k <= frame_length if OUTPUT_STE(i) == 1.0 o_arr1(j) = 0.2; else o_arr1(j) = -0.2; end k = k + 1; j = j + 1; end k = 1 + frame_overlap; end % output array for plot o_arr2 = zeros(length(A),1); j = 1;k = 1; for i = 1:no_of_frames while k <= frame_length if OUTPUT_ZCR(i) == 1.0 o_arr2(j) = 0.3; else o_arr2(j) = -0.3; end k = k + 1; j = j + 1; end k = 1 + frame_overlap; end x = linspace(1,length(A),length(A)); figure(1); plot(x,A) hold on plot(x,o_arr1,'-r') plot(x,o_arr2,'-g') legend('INPUT SIGNAL','OUTPUT WITH STE','OUTPUT WITH ZCR'); title('VOICED UNVOICED SEGMENTS') hold off % lpc for each frame (row wise) no_of_coeff = 13; %no of lpc coeff lpc_mat = zeros(no_of_frames, no_of_coeff); for i = 1:no_of_frames [lpc_coeff,g] = lpc(mat(i,:),no_of_coeff); lpc_mat(i,:) = lpc_coeff(2:end); end % excitation for voiced exciteV = zeros(1,frame_length); % (framelength) fs = Fs; %sampling freq is 8kHz pitch_period = 0.0075; %7.5 ms no_of_samples_in_one_period = fs*pitch_period; oneperiod = randi([0 1], 1,no_of_samples_in_one_period); %generating impulse of samples p = 0; l = frame_length/no_of_samples_in_one_period; for i = 1:l for j = 1:no_of_samples_in_one_period exciteV(1,p+j) = oneperiod(1,j); end p = p+no_of_samples_in_one_period; end % excitation for unvoiced exciteUV = normrnd(0,1,[1,frame_length]); synth_signal = zeros(1,length(A)); p=1; for i = 1:no_of_frames if OUTPUT_ZCR(i)==1 s = filter(1,[1 lpc_mat(i,:)], exciteV); else s = filter(1,[1 lpc_mat(i,:)], exciteUV); end synth_signal(p:p+frame_length-1)= s; p = p+frame_length; end %plot synthesised and original signal x = linspace(1,length(A),length(A)); figure(2); subplot(2,1,1); plot(x,A); title('SIGNAL PLOT') legend('original signal'); subplot(2,1,2); plot(x,synth_signal,'-r'); legend('synthesized signal'); %plot difference signal A = reshape(A,[1,length(A)]); rmse = sqrt(mean((A - synth_signal).^2)); figure(3); plot(x, A-synth_signal); title('DIFFERENCE SIGNAL') synth_signal = rescale(synth_signal,-1,1);