Applied Speech and Audio Processing: With matlab examples
Download 2.66 Mb. Pdf ko'rish
|
Applied Speech and Audio Processing With MATLAB Examples ( PDFDrive )
Speech communications
Figure 5.8 Use of LPC coefficients in an analysis filter. algorithms. As an example we can obtain the magnitude response of the synthesis fil- ter at N equally-spaced frequencies, and then determine the maximum resonance peak (perhaps corresponding to formant F1 in a speech sample), assuming an 8 kHz sample rate as follows: Fs=8000; %sample rate N=100; %frequency resolution [H, F] = freqz(1,a,N); %Plot the magnitude with a logarithmic scale on the y-axis semilogy(0:Fs/(N-1):Fs,abs(H)); [y,n]=max(abs(H)); PeakF=(n-1)*Fs/(N-1); The peak frequency of the plot is returned, along with the array index, by the max() function, and because Matlab indexes arrays from element 1, it is necessary to adjust the returned index, n, when determining the value in Hz. Both analysis and synthesis forms are used within LPC-based speech compression algorithms – one (or more) at either end. This is indicated in Figure 5.9 which shows a generic LPC-based speech coding algorithm. The same diagram, with some differences in how the excitation is derived or managed, probably represents all CELP class coders (which we will see later in Section 5.4.1), and also the GSM-style pulse-excitation coders (Section 5.3.1). In the encoder, original speech is normalised to be at some predetermined amplitude range, then the pitch information extracted to produce a residual. The residual, as men- tioned previously, contains vocal tract information which is modelled by LPC. Applying the LPC analysis filter to the residual will result in the vocal tract information being (mostly) removed, leaving a lung excitation signal which is modelled in some way and then transmitted to the decoder. The decoding process is essentially the reverse of the encoding process, and results in reconstructed speech – the fidelity of which depends on the ability of the algorithm to model the signals being parameterised, and the degree of quantisation applied to the parameters passed from encoder to decoder. Note that in the diagram of Figure 5.9 the extraction process (by which the various parameters are determined), has not been shown. Each of the parameters conveyed by the coder will be discussed in the sections which follow. |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling