Digital Speech Processing -- 2
Budget: $10 – $30 USD
Q1 answer by (BPFs) , Q2 Matlab
and see for the file.
Q1. In Table 1, 20 band-pass filters (BPFs) with their center frequencies and bandwidths are mentioned.
Suppose that these BPFs has triangular shape, where the center of a BPF has the magnitude 1 (one).
Table 1: Center frequencies and bandwidth of the filters (for speech data sampled at 16 kHz).
Find the output (magnitude) of your chosen filter. Show all the necessary steps of the calculation.
Q2. This is a programming assignment. Perform the following steps using MATLAB or any other
programming language.
a. Record and store your sustained /a/ voice for 2 seconds. The sampling frequency should be 8 KHz.
b. Divide the voice signal into frames of 30 milliseconds (ms) each. The frame rate is 10 ms.
c. Remove bias from the framed signal.
d. Apply pre-emphasis filter.
e. Apply Hamming window.
f. Apply FFT on the windowed signal.
g. Apply 24 BPFs (centered on the Mel scale). There will be a total of 24 outputs.
h. Apply discrete cosine transform (DCT) and retain the first 12 coefficients. These are MFCCs.
i. Calculate the power of the raw signal from each frame.
j. Find delta and double delta of the power and the MFCCs.
Show the outputs (in figures) of each step from b. to h. for the third frame. The figures should have
proper labels.
and see for the file.
Q1. In Table 1, 20 band-pass filters (BPFs) with their center frequencies and bandwidths are mentioned.
Suppose that these BPFs has triangular shape, where the center of a BPF has the magnitude 1 (one).
Table 1: Center frequencies and bandwidth of the filters (for speech data sampled at 16 kHz).
Find the output (magnitude) of your chosen filter. Show all the necessary steps of the calculation.
Q2. This is a programming assignment. Perform the following steps using MATLAB or any other
programming language.
a. Record and store your sustained /a/ voice for 2 seconds. The sampling frequency should be 8 KHz.
b. Divide the voice signal into frames of 30 milliseconds (ms) each. The frame rate is 10 ms.
c. Remove bias from the framed signal.
d. Apply pre-emphasis filter.
e. Apply Hamming window.
f. Apply FFT on the windowed signal.
g. Apply 24 BPFs (centered on the Mel scale). There will be a total of 24 outputs.
h. Apply discrete cosine transform (DCT) and retain the first 12 coefficients. These are MFCCs.
i. Calculate the power of the raw signal from each frame.
j. Find delta and double delta of the power and the MFCCs.
Show the outputs (in figures) of each step from b. to h. for the third frame. The figures should have
proper labels.
Related categories:
Engineering
Matlab and Mathematica
Electrical Engineering
Digital Signal Processing
MATLAB