基于MATLAB的男女声音分离源码.zip

%% 1.1 Read wav file and retrieve spectrum [y_m,Y_m, YA_m, YG_m] = spec('../dt/male.wav'); % Read audio of the male [y_f,Y_f, YA_f, YG_f] = spec('../dt/female.wav'); % Read audio of the female figure % Figure 1 plot (y_m); figure % Figure 2 plot (y_f); figure % Figure 3 imshow(YA_m); figure % Figure 4 imshow(YA_f); y_m=synth(Y_m); %sound(y_m,16000); % sound the re-synthesized audio %% 2.1 NMF traning [W_m,H_m]=cnmf(YA_m,20,500); % compute base matrix W; rr = 20; it = 500 figure % Figure 5 show_W(W_m); save ('W_m_20.mat','W_m'); %% 2.4 Getting resynthesized audio signal in time domain by Base1 [Y1_m]=maskspec(W_m,H_m,1,1); y1_m=synth(Y1_m.*YG_m); %sound(y1_m,16000); % sound the re-synthesized audio wavwrite(y1_m,16000,'base1_m.wav'); %% 2.5 Getting resynthesized audio signal in time domain by Base4 and Base8 [Y4_m]=maskspec(W_m,H_m,4,4); y4_m=synth(Y4_m.*YG_m); %sound(y4_m,16000); % sound the re-synthesized audio wavwrite(y4_m,16000,'base4_m.wav'); [Y8_m]=maskspec(W_m,H_m,8,8); y8_m=synth(Y8_m.*YG_m); %sound(y8_m,16000); % sound the re-synthesized audio wavwrite(y8_m,16000,'base8_m.wav'); %% 2.6 subplot and plot speech waveforms subplot(3,1,1); plot(y1_m); subplot(3,1,2); plot(y4_m); subplot(3,1,3); plot(y8_m); %% 2.7 try all the bases via different base number. Y_m_all=maskspec(W_m,H_m,1,20); y_m_all=synth(Y_m_all.*YG_m); %sound(y1_m,16000); % sound the re-synthesized audio wavwrite(y_m_all,16000,'base_20_m.wav'); [W_m_40,H_m_40]=cnmf(YA_m,40,500); % compute base matrix W; rr = 40; it = 500 figure % Figure 5 show_W(W_m_40); save ('W_m_40.mat','W_m_40'); Y_m_40=maskspec(W_m_40,H_m_40,1,40); y_m_40=synth(Y_m_40.*YG_m); %sound(y1_m,16000); % sound the re-synthesized audio wavwrite(y_m_40,16000,'base_40_m.wav'); [W_m_10,H_m_10]=cnmf(YA_m,10,500); % compute base matrix W; rr = 10; it = 500 figure % Figure 5 show_W(W_m_10); save ('W_m_10.mat','W_m_10'); Y_m_10=maskspec(W_m_10,H_m_10,1,10); y_m_10=synth(Y_m_10.*YG_m); %sound(y1_m,16000); % sound the re-synthesized audio wavwrite(y_m_10,16000,'base_10_m.wav'); [W_m_2000,H_m_2000]=cnmf(YA_m,2000,500); % compute base matrix W; rr = 10; it = 500 figure % Figure 5 show_W(W_m_2000); save ('W_m_2000.mat','W_m_2000'); Y_m_2000=maskspec(W_m_2000,H_m_2000,1,2000); y_m_2000=synth(Y_m_2000.*YG_m); %sound(y1_m,16000); % sound the re-synthesized audio wavwrite(y_m_10,16000,'base_10_m.wav'); %% 2.8 female video training [W_f,H_f]=cnmf(YA_f,20,500); % compute base matrix W; rr = 20; it = 500 figure % Figure 5 show_W(W_f); save ('W_f_20.mat','W_f'); %% Getting resynthesized audio signal in time domain by Base1 [Y1_f]=maskspec(W_f,H_f,1,1); y1_f=synth(Y1_f.*YG_f); %sound(y1_f,16000); % sound the re-synthesized audio wavwrite(y1_f,16000,'base1_f.wav'); %% Getting resynthesized audio signal in time domain by Base4 and Base8 [Y4_f]=maskspec(W_f,H_f,4,4); y4_f=synth(Y4_f.*YG_f); %sound(y4_f,16000); % sound the re-synthesized audio wavwrite(y4_f,16000,'base4_f.wav'); [Y8_f]=maskspec(W_f,H_f,8,8); y8_f=synth(Y8_f.*YG_f); %sound(y8_f,16000); % sound the re-synthesized audio wavwrite(y8_f,16000,'base8_f.wav'); %% subplot and plot speech waveforms subplot(3,1,1); plot(y1_f); subplot(3,1,2); plot(y4_f); subplot(3,1,3); plot(y8_f); %% try all the bases via different base number. Y_f_all=maskspec(W_f,H_f,1,20); y_f_all=synth(Y_f_all.*YG_f); %sound(y1_f,16000); % sound the re-synthesized audio wavwrite(y_f_all,16000,'base_20_f.wav'); [W_f_40,H_f_40]=cnmf(YA_f,40,500); % compute base matrix W; rr = 40; it = 500 figure % Figure 5 show_W(W_f_40); save ('W_f_40.mat','W_f_40'); Y_f_40=maskspec(W_f_40,H_f_40,1,40); y_f_40=synth(Y_f_40.*YG_f); %sound(y1_f,16000); % sound the re-synthesized audio wavwrite(y_f_40,16000,'base_40_f.wav'); [W_f_10,H_f_10]=cnmf(YA_f,10,500); % compute base matrix W; rr = 10; it = 500 figure % Figure 5 show_W(W_f_10); save ('W_f_10.mat','W_f_10'); Y_f_10=maskspec(W_f_10,H_f_10,1,10); y_f_10=synth(Y_f_10.*YG_f); %sound(y1_f,16000); % sound the re-synthesized audio wavwrite(y_f_10,16000,'base_10_f.wav'); [W_f_2000,H_f_2000]=cnmf(YA_f,2000,500); % compute base matrix W; rr = 10; it = 500 figure % Figure 5 show_W(W_f_2000); save ('W_f_2000.mat','W_f_2000'); Y_f_2000=maskspec(W_f_2000,H_f_2000,1,2000); y_f_2000=synth(Y_f_2000.*YG_f); %sound(y1_f,16000); % sound the re-synthesized audio wavwrite(y_f_2000,16000,'base_2000_f.wav'); %% 3.1 Clear the workspace clear all; clc; %% 3.2 Load base matrices load ('W_m_2000.mat'); load ('W_f_2000.mat'); %% 3.3 Load base matrices [y_mix,Y_mix, YA_mix, YG_mix] = spec('../dt/mix.wav'); % Read audio of the mix %% 3.4 get W matrics of W W_mix_2000 = [W_m_2000, W_f_2000]; [W_mix,H_mix]=cnmf(YA_mix,2000,500,false,W_mix_2000); %% 3.5 Compute YA_mix and separate speech [YA_m_from_mix]=maskspec(W_mix,H_mix,1,2000); y_m_from_mix=synth(YA_m_from_mix.*YG_mix); [YA_f_from_mix]=maskspec(W_mix,H_mix,2001,4000); y_f_from_mix=synth(YA_f_from_mix.*YG_mix); subplot(3,1,1); plot(y_mix); subplot(3,1,2); plot(y_m_from_mix); subplot(3,1,3); plot(y_f_from_mix); wavwrite(y_f_from_mix,16000,'f_speech_separated2000.wav'); wavwrite(y_m_from_mix,16000,'m_speech_separated2000.wav');