• Home
  • Textbooks
  • Audio Signal Processing and Coding
  • TRANSFORM CODERS

Audio Signal Processing and Coding

Andreas Spanias, Ted Painter, Venkatraman Atti

Chapter 7

TRANSFORM CODERS - all with Video Answers

Educators


Chapter Questions

Problem 1

Given the expressions for the DFT, the DCT, and the MDCT,
$$
\begin{aligned}
& X_{D F T}(k)=\frac{1}{\sqrt{2 M}} \sum_{n=0}^{2 M-1} x(n) e^{-j \pi n k / M}, 0 \leqslant k \leqslant 2 M-1 \\
& X_{D C T}(k)=c(k) \sqrt{\frac{2}{M}} \sum_{n=0}^{M-1} x(n) \cos \left[\frac{\pi}{M}\left(n+\frac{1}{2}\right) k\right], 0 \leqslant k \leqslant M-1
\end{aligned}
$$
where $c(0)=1 \sqrt{2}$, and $c(k)=1$ for $1 \leqslant k \leqslant M-1$
$$
\begin{aligned}
& X_{M D C T}(k)=\sqrt{\frac{2}{M}} \sum_{n=0}^{2 M-1} x(n) \underbrace{\sin \left[\left(n+\frac{1}{2}\right) \frac{\pi}{2 M}\right]}_{w(n)} \\
& \cos \left[\frac{(2 n+M+1)(2 k+1) \pi}{4 M}\right], \text { for } 0 \leqslant k \leqslant M-1
\end{aligned}
$$
FIGURE CANT COPY
Write the three transforms in matrix form as follows
$$
\mathbf{X}_{\mathbf{T}}=\mathbf{H x} \text {, }
$$
where $\mathbf{H}$ is the transform matrix, and $\mathbf{x}$ and $\mathbf{X}_{\mathrm{T}}$ denote the input and transformed vector, respectively. Note the structure in the transform matrices.

Check back soon!

Problem 2

Give the signal flowgraph of the FFT butterfly structure for an 8-point DFT, an 8-point DCT, and an 8-point MDCT. Specify clearly the values on the nodes and the branches. [Hint: See Problem 6.16 and Figure 6.18 in Chapter 6.]

Check back soon!

Problem 3

In this problem, we will study the energy compaction of the DFT and the DCT. Use $x(n)=e^{-0.5 n} \sin (0.4 \pi n), n=0,1, \ldots, 15$. Plot the 16-point DFT and 16-point DCT of the input signal, $x(n)$. See how the energy of the sequence is concentrated. Now pick two peaks of the DFT vector and the DCT vector and synthesize the input signal, $x(n)$. Let the synthesized signals be, $\hat{x}_{D F T}(n)$ and $\hat{x}_{D C T}(n)$. Compute the MSE values between the input signal and the two reconstructed signals. Repeat this for four peaks, six peaks, and eight peaks. Plot the estimated MSE values across the number of peaks selected and comment on your result.

Check back soon!

Problem 4

This computer exercise is a combination of Problems 2.24 and 2.25 in Chapter 2. In particular, the FFT analysis/synthesis module, in Problem 2.25, will be used within the two bands of the QMF bank. The configuration is shown in Figure 7.5.
a. Given, $H_0(z)=1-z^{-1}, H_1(z)=1+z^{-1}$. Choose $F_0(z)$ and $F_1(z)$ such that the aliasing term can be cancelled. Use $L=32$ and the peak-picking method for component selection. Perform speech synthesis and give timedomain plots of both input and output speech records.
b. Use the same voiced frame selected in Problem 2.24. Give time-domain and frequency-domain plots of $x_{d 0}^{\prime}(n)$ and $x_{d 1}^{\prime}(n)$ in Figure7.5.
c. Compute the overall SNR (between $x(n)$ and $x^{\prime}(n)$ ) and estimate a MOS score for the output speech.
d. Describe whether the perceptual quality of the output speech improves if the FFT analysis/synthesis module is employed within the subbands instead of using it for the entire band.

Check back soon!