Simulation 3

    

Figure: Synthetic vowel f₁(t).
Figure: Mixed signal f(t).

This simulation assumes that f₁(t) is a synthetic vowel as shown in Fig. , where F₀=125 Hz, N_F0=40, and it is a vowel /a/ synthesized by the LMA, and f₂(t) is a bandpassed random noise, where bandwidth of about 6 kHz. Three types of f(t) are used as simulation stimuli, where the SNRs of f(t) are from 0 to 20 dB in 10-dB steps. Mixed signal in case of SNR=10 dB is plotted in Fig. .

      

Figure: Precision for A_k(t) (SNR=10 dB).
Figure: SD for $\hat{f}_1(t)$ and the reduced SD of $\hat{f}_1(t)$.
Figure: Segregated signal $\hat{f}_1(t)$ (SNR=10 dB).

The simulations were carried out using the three mixed signals. The average SDs of f₁(t) and f(t), and the mean of the reduced SD of f₁(t) are shown in Fig. . Hence, it is possible to reduce the SD by about 15 dB as noise reduction, using the proposed method. For example, when the SNR of f(t) is 10 dB, the proposed method can segregate A_k(t) with a high precision as shown in Fig. , and can extract the $\hat{f}_1(t)$ shown in Fig. from the f(t) as shown in Fig. . Therefore, the proposed model can also extract the amplitude information of speech f₁(t) from a noisy speech f(t) with a high precision in which speech and noise exist in the same frequency region. Hence, this method can be applied in case where a speech signal is to be extracted from noisy speech.