Is it necessary to use filter bank in MFCC process? Can anyone explain what is the Mel filter? I know that frequency in hertz is converted into Mel scale but is this formula can be directly applied after the Fourier transformation of the speech signal? Why the filter bank is needed?
Asked
Active
Viewed 8,820 times
1 Answers
7
Mel filter bank is important due to following reasons:
It applies the Mel-frequency scaling, which is perceptual scale that helps to simulate the way human ear works. It corresponds to better resolution at low frequencies and less at high.
Using the triangular filter-bank helps to capture the energy at each critical band and gives a rough approximation of the spectrum shape, as well as smooths the harmonic structure. In theory you could manipulate on raw DFT bins, but then you are not reducing the dimensionality of your features - this is the whole point of doing filter-bank analysis, to capture the spectral envelope.
Obviously warping of frequency scale can be altered. Other mappings are possible, such as Bark, linear, etc. Mel is just one out of many options. To finalize, you can always calculate more filter-banks to have coarser scale (getting closer to actual DFT), but you will notice that at some point it will degrade the performance and your features do not carry essential information.
jojeck
- 11,107
- 6
- 38
- 74
-
Can you please explain it in simpler way.I am a Comp sc student and donot know much about signal processing – purple Dec 10 '14 at 18:45
-
@purple: You must therefore tell us which parts you don't understand. You don't understand DFT, perceptual scale, filter-bank? – jojeck Dec 10 '14 at 18:50
-
Fundamental of mfcc process and the filter bank concept.It would be great help – purple Dec 10 '14 at 18:53
-
These questions were already answered many times here: click, click, click, click. – jojeck Dec 10 '14 at 18:57
-
Can you explain what are fft bins – purple Dec 10 '14 at 19:06
-
That's what you get after performing DFT-complex values in frequency domain. Each sample corresponds to one bin separated by frequency resolution. – jojeck Dec 10 '14 at 19:12