Adaptive filter bank

Adaptive filter banks. In the basic configuration, all filter banks described above feature a time/frequency decomposition which is constant over time. As mentioned above, there are possibilities to switch the characteristics of a filter bank, going from one time/frequency decomposition to another one. We explain the basic principle using the example of MPEG Audio Layer 3 ... 

In an analysis/synthesis filter bank, all quantization errors on the spectral components show up on the time domain output signal as the modulated signal multiplied by the synthesis window. Consequently, the error is smeared in time over the length of the synthesis window / prototype filter. As described above, this may lead to audible errors if premasking is not ensured. This pre-echo effect (a somewhat misleading name, a better word would be pre-noise) can be avoided if the filter bank is not static, but switched between different frequency/time resolutions for different blocks of the overlap/add. An example of this technique called adaptive window switching is described below. 

As a further advantage of MDCT-like filter banks it should be noted that the time domain aliasing property needs to be valid for each half of the window independently from the other. Thus hybrid window forms (with different types of window functions for the first or second half) can be used. This leads to the realization of adaptive window switching systems ([Edler, 1989], see below). 

The first approach that need to be more precisely evaluated is the use of non-uniform filter banks [Petersen and Boll, 1981, Valiere, 1991], especially if they are applied in connection with perceptual criteria. Indeed, non-uniform filter banks allow a frequency dependent specification of the time-resolution/bandwidth compromise which could be adapted to the known features of our hearing system. The results of section 4.16 show that a high frequency-resolution is needed anyway, at least in the lower part of the spectrum, to ensure a sufficient separation of sinusoidal signal components from the noise. 

Forward- or backward adaptive predictors can be used to increase the redundancy removal capability of an audio coding scheme. In the case of high resolution filter banks backward adaptive predictors of low order have been used with success [Fuchs, 1995],... 

In summary it can be stated that hybrid filter banks allow increased flexibility in the design by including the possibility to have different frequency resolutions at different frequencies. Another degree of freedom which is gained by using hybrid filter banks is the adaptive switching of the filter bank to different time/frequency behavior. On the downside, a somewhat increased complexity compared to solutions based on cosine-modulated filter banks is necessary to implement adaptive hybrid systems. 

For high frequency resolution filter bank based coders the transmission of prediction coefficients would take a huge amount of additional side information. Therefore, a short (two tap) backward adaptive predictor is used. An attenuation factor is applied to the predictor to lower the long term impact of a connection loss. Prediction is switched on and off to ensure it is only used in blocks with an actual prediction gain. Additionally, the predictors are reset in certain intervals. In this way, small differences in the arithmetic accuracy between encoder and decoder do not lead to audible errors and can be tolerated. 

To adapt original TDNN system for the vowels recognition in children, some modifications are made. In this study, there will be six vowels compared to three consonants used by Waibel. Basically this TDNN and the original TDNN have the same number of layers one input layer, two hidden layers and one output layers. At the input layer, this study uses 24 units of cepstral coefficient instead of 16 melscale filter bank coefficients. However, the sizes of the window frames are the same between these two architectures. The proposed TDNN uses 15 window frames with different... 

Space requirement 1. Well adapted for low headroom requirements 2. Form of banks can be chosen to fit any shaped space 3. Space should be allowed for a man to remove filter cells for cleaning or replacement 1. Have a high headroom requirement 2. Take up less floor space than other types... 

---