| New high capacity optical discs and high bandwidth networks provide the capability for delivering multichannel audio. Although there are many one- and two-channel recordings, only a handful of multichannel recordings exist. Hence, we propose neural network approaches to synthesize microphone signals with the correct acoustical characteristics of specific venues. In this research, the complex semi-cepstrum technique is employed to extract features from music signals, whereas the fuzzy cerebellar model articulation controller (FCMAC) is used for training the signals. An alternative approach based on the nonuniform rational B-spline (NURBS) is also proposed for spectrogram morphing. Applications of these methods include remastering of existing monophonic and stereophonic recordings for multichannel rendering.; Another work focuses on a systematic treatment for developing a noise removal system based on the fundamental principle of reinforcement learning and FCMAC algorithm. The proposed system improves its performance over time through two mechanisms. First, the modified stochastic real-valued (SRV) algorithm, learning from its own mistakes via the reinforcement signal and reinforcing its action to improve future performance, is used for searching the optimal noise frequencies for the overall training system. Second, system states associated with the positive reinforcement are memorized by the FCMAC-based neurons where in the future, similar states will share the experiences already stored there and then lead the action to a more positive situation.; In order to achieve a better performance, the above works are integrated with the rational filter bank, a general structure for two-band filter banks with arbitrary sampling factors. This technique extends the quadrature mirror filter (QMF) bank to fit all spectrum-separation cases. The spectral analysis shows that the original band of the input signal can be unevenly decomposed into a given ratio and then reconstructed with negligible distortion. To achieve high precision filter banks and reduce the implementation cost, we also integrate the rational filter banks with the frequency response masking (FRM) technique. This procedure can be applied to filter banks with both integer and rational decimation factors. Comparison with conventional M-stage octave filter banks is also given to illustrate the advantages of the proposed system. |
