Need to design a filter to allow only voice frequencies

[sundar_milo]

Hi.
I need a bandpass filter that should be able to allow only our voice frequencies and reject all other frequencies.I dont have any idea about the frequency range of voice but  i think that is not a problem at all.I searched for BPF in google and found this :

http://www.electronics-tutorials.com/filters/active-bandpass-filters.htm

it will be easy for me to deal with Opamp based filters rather than others that use inductors.

My input to the filter will be thru' a mic and an amplifier . Is this filter good for my project or is there any other filter that well suits my project?

Also i doubt if the frequency range that i want to design is feasible.

Any help would be much appreciated.Thanks.

 

[audioguru2]

Hi Sundar,
The Multiple Feedback Bandpass Filter circuit you posted passes a very narrow range of frequencies. "Telephone Quality" voice frequencies are from 300Hz to 3kHz.
You need a highpass filter to reject frequencies lower than 300Hz followed by a lowpass filter to reject frequencies higher than 3kHz.

Try this circuit:
http://www.radiolocman.com/electrical-engineering/circuit-cache.html?di=11095

 

[Yoyok]

Or try this, www.ti.com/sc/docs/apps/msp/journal/nov2000/nov_08.pdf.

 

[sundar_milo]

Thanks for those replies.Now im confused with these two circuits given in the links you have posted.But i would like to go with the one that has a single supply version.According to the circuit given in the link provided  by Audioguru , it has to be fed with a dual supply(+15V and -15V).However a single supply can be converted to a dual supply , so there wont be much problem at all.

The PDF file from yoyok has a half-supply generator that forms a virtual ground.So im planning now to use the ckt from Audioguru and feed it with power supply using the half-supply generator.

 

[audioguru2]

The circuit posed by me will work fine when both opamps are fed from the half-supply generator.

 

[MP1]

sundar,
You did not mention how much attenuation/rejection is needed. If you need to eliminate the frequencies above and below these set points, you will need a high order filter. Neither of these designs are high order. They will give you some attenuation, but not rejection. Will this work for you? You did not mention much about the purpose of the design.

MP

 

[sundar_milo]

Hi MP. I want to design a voice recognition system using TMS 320 C54X  DSP processor.To which im going to give the input voice signal through mic,amplifier and filter sections.

First im concentrating in designing a good filter for this project.I dunno the attenuation and rejection values in dB for our voice signals .Even i came to know of the frequency range 300 Hz to 3 KHz from the above posts only.But i would like to know how to estimate these values for my project.

Any help will be appreciated.Thank you very much.

 

[mvs sarma]

hi sundar_milo

A filter alone cont, i fear, identify whether it is voice. it can only decide whether the frequencies are within the so called voice band(as per  the term  generally used in telecom  .3 to 3.4 KHz and in some countires .3 to 3 KHz)

now if the signal contains signals like music tones like DTMF within the pass band of these filters they sicerely allow.

the characteristic feature of voice is that it is not continuous  and there are define breaks. perhaps your algorithm should also check for this.

sarma

 

[MP1]

sundar,
The filter that was posted by audioguru will give you more attenuation than the bandpass filter posted. The reason bandpass will not give you the results you want is that Q is inversely proportional to Bandwidth. As you increase the bandwidth of a bandpass filter (300 to 3K is a wide bandwidth), you lose the Q, which directly effects the amount of cut or boost in a filter.
The Sams circuit posted above is a standard speech filter used in many applications. I have used the same schematic, only different resistor and capacitor values.
To get an even better cut point, you can build high order low pass and high pass filters which are in the 3rd or 4th order and just connect them together like in the Sams design.

As sarma has mentioned, once you get rid of the frequencies that do not hold voice, how will you determine what is voice content and what is something else in the 300 to 3K range? Just curious.

MP

 

[sundar_milo]

Thanks for sharing those info.i am planning to use an algorithm for correlation between predefined recorded signal and the input voice signal and if it is closer to 1 then that instruction is executed whatever it has been programmed for.

Just like the dot product of two vectors a.b=1 if they are parallel and a.b=0 if they are perpendicular, the correlation value is similar to that.

Btw can u explain more on the attenuation part ? like the values that are  needed to make a good filter .And im sure there is no way to filter all the signals other than the voice signals,but atleast it should contain only the voice band so that correlation is easier to perform and not tideous one.

 

[mvs sarma]

Hi
perhaps the attenuation out side the audio band  depends on the % tolerence of the components you use across the filter like the one suggested by Audioguru

Try toassmble one and  run a test with a set of frequencies,say , from 200Hz to 4 or 5KHz  and tabulate you will know the filter behaviour. 

you may have the set up at any lab near to you.

sarma

 

[audioguru2]

The article posted by Yoyok shows how much better a 4th order filter attenuates outside the bandpass than a 2nd order filter. The circuit posted by me is a 2nd order filter but is inverting.

View attachment 39960

 

[sundar_milo]

A 4th order filter can do better in my project.First i shall design one and test it with different frequencies , plot the gain vs freq.

 

[MP1]

Here is an interesting project done at cornell university:

http://instruct1.cit.cornell.edu/courses/ee476/FinalProjects/s2006/XL76_SL362/XL76%20SL362/index.html


MP