Voice Recognition 2 Question

I have found a code in voice recognition. http://crunchmodo.com/digital-signal-processing-voice-recognition-in-matlab-source-code/

1.-what does the y axis and x axis represent? Does y represent the value of data and x the number of samples? 2.-why does he use abs? to make the signal positive? if yes why he want to make it positive? it is more efficient to compare two sound file in positive value?

Answer 1

[Walter Roberson]

X represents the number of samples. The Y is the intensity at that sample number.

What does class(usertemp) show after you run the code? If it shows anything other than 'uint8' then the intensities returned by wavrecord() are both positive and negative and abs() should not be used. If it is 'uint8' then the intensities returned are only positive and the abs() does not change anything. I would therefore remove the abs() for both the wavread() and wavrecord(). The comment about cropping to the part that contains speech is incorrect.

Answer 2

[JOJO]

1.-I plot the graph, it show that the signal returned is positive and negative by the wavrecord(). I also ploted the signal returning from abs, the signal returning from abs is only positive.

2.-Why abs should not be used?

3.-it is more efficient to compare two sound file in positive value?

4.-I tested comparing in positive value it worked, if I do it in postive and negative it would be better?

[Walter Roberson]

The data being read is the distance above or below the middle intensity unless you are using uint8. Quiet sounds are below the middle intensity. abs() on the data is not going to clip the data (like the comment says), it is going to transform the very quiet data to look like it is very loud data, which is going to give you a very misleading idea of the sound.

It is not more efficient to compare tw0 sounds by only considering the positive values: it will be more inaccurate. And it will sound wrong when you play it.