Conclusions

This paper proposed a method of extracting the desired signal from a noisy signal, addressing the problem of segregating two acoustic sources as a model of acoustic source segregation based on Auditory Scene Analysis. The problem of segregating two acoustic sources, discussed here, is an ill inverse problem in which the instantaneous amplitude and the instantaneous phase of the desired signal must be determined using the instantaneous amplitude and the instantaneous phase of the observed signal. This method uses the instantaneous amplitude and the instantaneous phase of signal component passed through a wavelet filterbank. It can solve this problem using constraints, which are related to the four heuristic regularities proposed by Bregman.

As an example of segregation using the proposed method, we demonstrated three simulations of segregating two acoustic sources. These simulations were:

1.

Extracting an AM complex tone from a noise-added AM complex tone.

2.

Extracting one AM complex tone from mixed AM complex tones.

3.

Extracting a speech signal from a noisy speech.

In these simulations, two measures were used to evaluate the proposed method. One was precision, which is the temporal average of the segregated error for A_k(t) and the other was the spectrum distortion for the extracted signal. Moreover, the segregation accuracy between using the proposed model and using the other method, where it uses the omitted constraints in the proposed model, were evaluated by computer simulations.

The results of simulations 1 and 2 showed that the proposed method could extract with high precision the AM complex tone not only from a noise-added AM complex tone but also from mixed AM complex tones, in which signal and noise existed in the same frequency region. In particular, it was possible to reduce the SD by about 20 dB as noise reduction, using the proposed method. Moreover, the results of simulation 3 showed that the proposed method could also extract the speech signal from noisy speech.

Comparing the proposed method with the three conditions, we found that using it with constraints related to the four regularities is better than using the other method (under three conditions). In particular, it could segregate the desired signal from the mixed signal, with about 4 dB better precision and about 10 dB improvement in spectrum distortion.