filtering question

Hi all, i have a question on filtering here.apparently my filter plot shows
it can get rid of frequency up to 0.017. here's the code,just cut and
paste.

%--------------------------------------------------------------
Wp = 0.0295;
Ws = 0.0270;
Rp = 1;
Rs = 32;
[N,Wn] = cheb1ord(Wp,Ws,Rp,Rs);
[b,a] = cheby1(N,Rp,Wn,'high');
 [h,omega] = freqz(b,a,80);
 plot (omega/pi,(abs(h)));grid;
 xlabel('\omega/\pi'); ylabel('Magnitude');
 title('Type I Chebyshev Highpass Filter');
%--------------------------------------------------------------
but i pass a signal through it and plot the filtered signal, i took the
noisyspeech in a frame,80 samples. do a 

y=filter(b,a,noisyspeech_in frame1);
Y=abs(fft(y));
f=f=(0:length(Noi)-1)/2/0.005;
plot(f,Y)

it still shows components between 0 to 60Hz. Why is this so??

>Hi all, i have a question on filtering here.apparently my filter plot
shows
>it can get rid of frequency up to 0.017. here's the code,just cut and
>paste.
>
>%--------------------------------------------------------------
>Wp = 0.0295;
>Ws = 0.0270;
>Rp = 1;
>Rs = 32;
>[N,Wn] = cheb1ord(Wp,Ws,Rp,Rs);
>[b,a] = cheby1(N,Rp,Wn,'high');
> [h,omega] = freqz(b,a,80);
> plot (omega/pi,(abs(h)));grid;
> xlabel('\omega/\pi'); ylabel('Magnitude');
> title('Type I Chebyshev Highpass Filter');
>%--------------------------------------------------------------
>but i pass a signal through it and plot the filtered signal, i took the
>noisyspeech in a frame,80 samples. do a 
>
>y=filter(b,a,noisyspeech_in frame1);
>Y=abs(fft(y));
>f=f=(0:length(Noi)-1)/2/0.005;
>plot(f,Y)
>
>it still shows components between 0 to 60Hz. Why is this so??
>
>
>
sorry, it should be length(Y) instead of length(Noi)

>Hi all, i have a question on filtering here.apparently my filter plot
shows
>it can get rid of frequency up to 0.017.
>
>it still shows components between 0 to 60Hz. Why is this so??

Hi,

Maybe you should not look at it with two different frequency scales.
Your plot of the filter characteristic is on a normalized frequency
scale.
So if you plot both with normalized frequency
scale, you will see that it works as expected.

gr.
Anton

>>Hi all, i have a question on filtering here.apparently my filter plot
>shows
>>it can get rid of frequency up to 0.017.
>>
>>it still shows components between 0 to 60Hz. Why is this so??
>
>Hi,
>
>Maybe you should not look at it with two different frequency scales.
>Your plot of the filter characteristic is on a normalized frequency
>scale.
>So if you plot both with normalized frequency
>scale, you will see that it works as expected.
>
>gr.
>Anton
>
>
>
>

oh gosh, now i tried to run my filter again and it is not zero at 0.017
anymore. Does the filter differ in every run? I tried putting Ws=0.019 and
Wp=0.0225, to achieve my objective of having zero component at 0.017 but
the lower cutoff does not change at all! It remains below 0.017. Did i do
something wrong?

Moreover,i have tried to plot them on the same scale but there is still
some coponent at 0. Why cant i get rid of them?

Thanks for your help.

"doggie" <elusivetruelove2003@yahoo.com> wrote in message 
news:l6idnRp1OsyT3fLZRVn-tg@giganews.com...
> Hi all, i have a question on filtering here.apparently my filter plot 
> shows
> it can get rid of frequency up to 0.017. here's the code,just cut and
> paste.
>
> %--------------------------------------------------------------
> Wp = 0.0295;
> Ws = 0.0270;
> Rp = 1;
> Rs = 32;
> [N,Wn] = cheb1ord(Wp,Ws,Rp,Rs);
> [b,a] = cheby1(N,Rp,Wn,'high');
> [h,omega] = freqz(b,a,80);
> plot (omega/pi,(abs(h)));grid;
> xlabel('\omega/\pi'); ylabel('Magnitude');
> title('Type I Chebyshev Highpass Filter');
> %--------------------------------------------------------------
> but i pass a signal through it and plot the filtered signal, i took the
> noisyspeech in a frame,80 samples. do a
>
> y=filter(b,a,noisyspeech_in frame1);
> Y=abs(fft(y));
> f=f=(0:length(Noi)-1)/2/0.005;
> plot(f,Y)
>
> it still shows components between 0 to 60Hz. Why is this so??

I wonder if the problem might be that 80 samples is too short to get such 
fine frequency resolution in the result?

Fred 

Hi,

Of course I don't have your testfile, but I gave it a
short look. I tested the filter with white noise (1024 samples). 
But with a much shorter length (like 50 samples), I don't get
a good result.

gr.
Anton 

>Hi,
>
>Of course I don't have your testfile, but I gave it a
>short look. I tested the filter with white noise (1024 samples). 
>But with a much shorter length (like 50 samples), I don't get
>a good result.
>
>gr.
>Anton 
>
>

Hi,what i'm trying to do is to filter each frame of speech (10ms) so that
there is no more dc component and 60Hz hum. This will ensure i will
measure the correct zero crossing rate subsequently. So what do you
suggest i do? If i filter the whole length of speech instead, the
individual frame might not have zero dc and 60 hz component. Pls advise.

"doggie" <elusivetruelove2003@yahoo.com> wrote in message 
news:-NSdnTMEXsWVzezZnZ2dneKdnZydnZ2d@giganews.com...
> >Hi,
>>
>>Of course I don't have your testfile, but I gave it a
>>short look. I tested the filter with white noise (1024 samples).
>>But with a much shorter length (like 50 samples), I don't get
>>a good result.
>>
>>gr.
>>Anton
>>
>>
>
> Hi,what i'm trying to do is to filter each frame of speech (10ms) so that
> there is no more dc component and 60Hz hum. This will ensure i will
> measure the correct zero crossing rate subsequently. So what do you
> suggest i do? If i filter the whole length of speech instead, the
> individual frame might not have zero dc and 60 hz component. Pls advise.

to get 60Hz resolution you need 1/60 secs long filter.

To get a 60Hz highpass with sharper (say 1Hz) transition, you need a filter 
that is 1 second long.

Then, using either filter, there is a transient response that is as long as 
the filter - so you don't get the steady state response unless you process 
data that is longer....

Fred 

>
>"doggie" <elusivetruelove2003@yahoo.com> wrote in message 
>news:-NSdnTMEXsWVzezZnZ2dneKdnZydnZ2d@giganews.com...
>> >Hi,
>>>
>>>Of course I don't have your testfile, but I gave it a
>>>short look. I tested the filter with white noise (1024 samples).
>>>But with a much shorter length (like 50 samples), I don't get
>>>a good result.
>>>
>>>gr.
>>>Anton
>>>
>>>
>>
>> Hi,what i'm trying to do is to filter each frame of speech (10ms) so
that
>> there is no more dc component and 60Hz hum. This will ensure i will
>> measure the correct zero crossing rate subsequently. So what do you
>> suggest i do? If i filter the whole length of speech instead, the
>> individual frame might not have zero dc and 60 hz component. Pls
advise.
>
>to get 60Hz resolution you need 1/60 secs long filter.
>
>To get a 60Hz highpass with sharper (say 1Hz) transition, you need a
filter 
>that is 1 second long.
>
>Then, using either filter, there is a transient response that is as long
as 
>the filter - so you don't get the steady state response unless you
process 
>data that is longer....
>
>Fred 
>
>
>

ok..but i can't increase the length of my speech frame if not the short
term stationery theory may not stand. What should i do? Im being stuck at
this step for so long. My main aim is just to get rid of dc component and
50/60 Hz hum so i can calculate zero crossing rate accurately. I tried
filtering the whole length of speech but when seperated into individual
frames, it still has dc and 60Hz components so i have to do it frame by
frame.

By the way, we cant get rid of these components completely,can we? 

Please advise what filter i should use as im really not strong in filter
design. I just need to get over this part and my algorithm will be
complete. Thanks for the advices.

doggie wrote:

> 
> ok..but i can't increase the length of my speech frame if not the short
> term stationery theory may not stand. What should i do? Im being stuck at
> this step for so long. My main aim is just to get rid of dc component and
> 50/60 Hz hum so i can calculate zero crossing rate accurately. I tried
> filtering the whole length of speech but when seperated into individual
> frames, it still has dc and 60Hz components so i have to do it frame by
> frame.
> 


If the filtering of the "whole length" did the job then the existence of
those frequencies in the frame is simply an artifact of chunking it up
into frames. Your  "calculate zero crossing rate accurately" will be
similarly affected - moving a frame just a little could easily change
the number of zero crossings per frame. All this means is that the frame
boundaries are chosen in some arbitrary manner that has no direct
relationship to the content of the signal.
	Just out of curiosity what does an accurate measurement of zero
crossings per frame tell you about a speech record? Pardon if you've
already explained it I haven't been following closely.

-jim



> By the way, we cant get rid of these components completely,can we?
> 
> Please advise what filter i should use as im really not strong in filter
> design. I just need to get over this part and my algorithm will be
> complete. Thanks for the advices.

----== Posted via Newsfeeds.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ Newsgroups
----= East and West-Coast Server Farms - Total Privacy via Encryption =----