Impulse/noise models

Despite its ability to faithfiilly mimic the target and transition patterns of natural speech, standard LP synthesis has a significant unnatural quality to it, often impressionisticly described as buzzy or metallic sounding. Recall that while we measured the vocal tract model parameters directly from real speech, we still used an explicit impulse/noise model for the source. As we will now see, it is this, and specifically the interaction of this with the filter, which creates the unnaturalness. 

Autoregressive (AR) model-based Click Detection. In this method ([Vaseghi and Rayner, 1988, Vaseghi, 1988, Vaseghi and Rayner, 1990]) the underlying audio data. v n is assumed to be drawn from a short-term stationary autoregressive (AR) process (see equation (4.1)). The AR model parameters a and the excitation variance <52e are estimated from the corrupted data x[n using some procedure robust to impulsive noise, such as the M-estimator (see section 4.2). 

MFCC synthesis is a technique which attempts to S3mthesise from a representation that we use because of its statistical modelling properties. A completely accurate S3mthesis from this is not possible, but it is possible to perform fairly accurate vocal tract filter reconstruction. Basic techniques use an impulse/noise excitation method, while more advanced techniques attempt a complex parameterisation of the source. 

The source for vowels and voiced consonants can be generated either as an explicit time-domain periodic function, of the type described in Section 11.4.2, or by an impulse sequence that is fed into a glottal LTI filter. For obstruent sounds a random-noise generator is used. Sounds such as voiced fricatives use both sources. If we adopt the impulse/filter approach, we see that in fact these sources are equivalent to those of the classical impulse/noise LP model (Section 12.6.4). One difference between the impulse/glottal-filter source and LP models is that in the impulse/glottal-filter model the voiced source is... 

Coloured noise, narrow band noise and periodic impulsive noise are usually modeled as background noise because they remain stationary from seconds to even hours. Periodic impulsive noise synchronous to the mains and asynchronous impulsive noise may cause bit or burst errors over the transmission, although they are time variant. A complete theoretical analysis on shown noises in Figure 2 can be found on [5, 6,11,12,13]. 

Fig. 10. Impulsive noise obtained within the scenario 3 a) easel, b) case2 The impulsive noise can be modeled by the equation (6)...

Zimmermann, M. Dostert, K "Analysis and modeling of impulsive noise in broad-band... 

D. Umehara, S. Hirata, S. Denno, and Y. Morihiro, "Modeling of Impulse Noise for... 

---