Applied Speech and Audio Processing: With matlab examples
Download 2.66 Mb. Pdf ko'rish
|
Applied Speech and Audio Processing With MATLAB Examples ( PDFDrive )
- Bu sahifa navigatsiya:
- 2.6.2.1 Correlogram
2.6. Visualisation
27 Due to the large dynamic range of audio signals, it is common to plot the logarithm of absolute amplitude, rather than the amplitude directly; thus the spectrum will often be plotted as a power spectrum, using 20 × log 10 (spectrum), or with the Matlab: semilogy(spectrum); The x-axis, showing frequency, is generally plotted linearly. The labelling of this axis defaults to the index number of the bins in the FFT output spectrum – not particularly useful in most cases. A far better approach is to specify this yourself either directly in Hz, scaled between 0 and 1 (DC to Nyquist frequency), or in radians between 0 and π. res=pi/size(spectrum); semilogy(res:res:pi, spectrum); This radian measure, normally denoted by the independent frequency variable being written as ω, represents 2π as the sampling frequency, 0 as DC and thus π to be the Nyquist frequency. It is referred to as angular frequency or occasionally natural frequency, and considers frequencies to be arranged around a circle. The notation is useful when dealing with systems that are over- or undersampled, but apart from this, it is more consistent mathematically because it means equations can be derived to describe a sampled system that do not depend on the absolute value of sample rate. 2.6.2 Other visualisation methods As you may expect, many other more involved visualisations exist, and which have evolved as being particularly suited for viewing certain features. One of the most useful for speech and audio work is the linear prediction coefficient spectral plot that will be described in Section 5.2.1. Here, on the other hand, two very general and useful methods are presented – namely the correlogram and the cepstrum. 2.6.2.1 Correlogram A correlogram is a plot of the autocorrelation of a signal. Correlation is the process by which two signals are compared for similarities that may exist between them either at the present time or in the past (however much past data are available). Mathematically it is relatively simple. We will start with an equation, defining the cross-correlation between two vectors x and y performed at time t, and calculating for the past k time instants, shown in Equation (2.4): c x,y [t] = k x [k]y[t − k]. (2.4) In Matlab such an analysis is performed using the xcorr function over the length of the shortest of the two vectors being analysed: |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling