Formant filter

  • Author or source: Alex
  • Created: 2002-08-02 18:26:59
code
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
/*
Public source code by [email protected]
Simple example of implementation of formant filter
Vowelnum can be 0,1,2,3,4 <=> A,E,I,O,U
Good for spectral rich input like saw or square
*/
//-------------------------------------------------------------VOWEL COEFFICIENTS
const double coeff[5][11]= {
{ 8.11044e-06,
8.943665402,        -36.83889529,   92.01697887,    -154.337906,    181.6233289,
-151.8651235,   89.09614114,        -35.10298511,   8.388101016,    -0.923313471  ///A
},
{4.36215e-06,
8.90438318, -36.55179099,   91.05750846,    -152.422234,    179.1170248,  ///E
-149.6496211,87.78352223,   -34.60687431,   8.282228154,    -0.914150747
},
{ 3.33819e-06,
8.893102966,        -36.49532826,   90.96543286,    -152.4545478,   179.4835618,
-150.315433,        88.43409371,    -34.98612086,   8.407803364,    -0.932568035  ///I
},
{1.13572e-06,
8.994734087,        -37.2084849,    93.22900521,    -156.6929844,   184.596544,   ///O
-154.3755513,       90.49663749,    -35.58964535,   8.478996281,    -0.929252233
},
{4.09431e-07,
8.997322763,        -37.20218544,   93.11385476,    -156.2530937,   183.7080141,  ///U
-153.2631681,       89.59539726,    -35.12454591,   8.338655623,    -0.910251753
}
};
//---------------------------------------------------------------------------------
static double memory[10]={0,0,0,0,0,0,0,0,0,0};
//---------------------------------------------------------------------------------
float formant_filter(float *in, int vowelnum)
{
                res= (float) ( coeff[vowelnum][0]  *in +
                                     coeff[vowelnum][1]  *memory[0] +
                                     coeff[vowelnum][2]  *memory[1] +
                                     coeff[vowelnum][3]  *memory[2] +
                                     coeff[vowelnum][4]  *memory[3] +
                                     coeff[vowelnum][5]  *memory[4] +
                                     coeff[vowelnum][6]  *memory[5] +
                                     coeff[vowelnum][7]  *memory[6] +
                                     coeff[vowelnum][8]  *memory[7] +
                                     coeff[vowelnum][9]  *memory[8] +
                                     coeff[vowelnum][10] *memory[9] );

memory[9]= memory[8];
memory[8]= memory[7];
memory[7]= memory[6];
memory[6]= memory[5];
memory[5]= memory[4];
memory[4]= memory[3];
memory[3]= memory[2];
memory[2]= memory[1];
memory[1]= memory[0];
memory[0]=(double) res;
return res;
}

Comments

Where did the coefficients come from? Do they relate to frequencies somehow? Are they male or female? Etc.
And are the coeffiecients for 44k1hz?
/stefancrs
It seem to be ok at 44KHz although I get quite lot of distortion with this filter.
There are typos in the given code too, the correct version looks like this i think:
float formant_filter(float *in, int vowelnum)
{
  float res= (float) ( coeff[vowelnum][0]* (*in) +
  coeff[vowelnum][1] *memory[0] +
  coeff[vowelnum][2] *memory[1] +
  coeff[vowelnum][3] *memory[2] +
  coeff[vowelnum][4] *memory[3] +
  coeff[vowelnum][5] *memory[4] +
  coeff[vowelnum][6] *memory[5] +
  coeff[vowelnum][7] *memory[6] +
  coeff[vowelnum][8] *memory[7] +
  coeff[vowelnum][9] *memory[8] +
  coeff[vowelnum][10] *memory[9] );
...

(missing type and asterisk in the first calc line ;).

I tried morphing from one vowel to another and it works ok except in between 'A' and 'U' as I get a lot of distortion and sometime (depending on the signal) the filter goes into auto-oscilation.

Sebastien Metrot
How did you get the coeffiecients?

Did I miss something?

/Larsby
Yeah, morphing lineary between the coefficients works just fine. The distortion I only get when not lowering the amplitude of the input. So I lower it :)

Larsby, you can approximate filter curves quite easily, check your dsp literature :)
Correct, it is for sampling rate of 44kHz.
It supposed to be female (soprano), approximated with its five formants.

--Alex.
Can you tell us how you calculated the coefficients?
The distorting/sharp A vowel can be toned down easy by just changing the first coeff from 8.11044e-06 to 3.11044e-06. Sounds much better that way.
Hi, I get the last formant (U) to self-oscillate and distort out of control whatever I feed it with. all the other ones sound fine...

any sugesstions?

Thanks,
Jonas
I was playing around this filter, and after hours of debugging finally noticed that converting those coeffecients to float just won't do it. The resulting filter is not stable anymore. Doh...

I don't have any idea how to convert them, though.
  • Date: 2008-10-29 00:35:27
  • By: mysterious T
Fantastic, it's all I can say! Done the linear blending and open blending matrix (a-e, a-i, a-o, a-u, e-i, e-o...etc..etc..). Too much fun!

Thanks a lot, Alex!
  • Date: 2010-12-14 13:16:19
  • By: johnny
What about input and output range? When I feed the filter with audio data in -1 to 1 range, output doesn't stay in the same range. Maybe the input or output needs to be scaled?