Re: How can I increase/decrease (frequency/pitch) and phase using fft/ifft tia sal22

[Sergei Steshenko]

--- On Thu, 4/7/11, Rick T <address@hidden> wrote:

From: Rick T <address@hidden>
Subject: Re: How can I increase/decrease (frequency/pitch) and phase using 
fft/ifft tia sal22
To: address@hidden
Date: Thursday, April 7, 2011, 7:43 AM

Greetings All
I went back and did what Sergei recommend and used resample and repmat, but I'm 
noticing that on some of the values the rows aren't the same as the sample 
rate, see image link below.  notice the top image value for rows says 1000 and 
the bottom image says rows = 1008.  This happens when I change the values of 
resample and repmat (freq_new) but only for certain values.  How can I fix this 
correctly?  I could just delete everything after 1000 but I'm not sure if this 
is a bug or just the way resample/repmat works.  PS: I'm using octave 3.2.4


http://dl.dropbox.com/u/6576402/questions/rows_different.png
Here's the test code I used to test this



#yiv21956316 p, #yiv21956316 li {white-space:pre-wrap;}

%resample_repmat signal
clear all, clf

Fs = 1000; % Sampling rate
Ts = 1/Fs; %sampling interval
t=0:Ts:1-Ts; %sampling period




freq_orig=1;
y=sin(2*pi*t*freq_orig)'; %gives a short wave




freq_new=9;
y2=resample(y,1,freq_new); %resample matrix
y3=repmat (y2,freq_new,1); %replicate matrix




[r_orig,c_orig] = size(y) %get orig number of rows and cols
[r_new,c_new] = size(y3) %get new number of rows and cols




subplot(2,1,1),plot(y),title('Orginal signal')
title(['rows=',num2str(r_orig),' cols=',num2str(c_orig)])
subplot(2,1,2),plot(y3),title('New signal')
title(['rows=',num2str(r_new),' cols=',num2str(c_new)])


On Mon, Apr 4, 2011 at 1:17 PM, Sergei Steshenko <address@hidden> wrote:




--- On Mon, 4/4/11, Rick T <address@hidden> wrote:



From: Rick T <address@hidden>

Subject: Re: How can I increase/decrease (frequency/pitch) and phase using 
fft/ifft tia sal22

To: "Sergei Steshenko" <address@hidden>

Cc: address@hidden

Date: Monday, April 4, 2011, 3:37 PM



I need fft/ifft due to the fact that I have to alter various cells in the array 
the signal is stored in (in the frequency domain.  The script is very long.  In 
my experience if you post hundreds of lines of code people will not look at it. 
 That's why I kept it simple and basic.  And asked How can I increase/decrease 
(frequency/pitch) and phase using fft/ifft. 









PS: Unfortunately Sergei the nice solution you sent won't work, I'm dealing 
with large arrays that are exported back out as audio files and fft/ifft seems 
to be the fastest.

 







On Mon, Apr 4, 2011 at 11:33 AM, Sergei Steshenko <address@hidden> wrote:









--- On Mon, 4/4/11, Rick T <address@hidden> wrote:







From: Rick T <address@hidden>



Subject: How can I increase/decrease (frequency/pitch) and phase using fft/ifft 
tia sal22



To: address@hidden



Date: Monday, April 4, 2011, 2:11 PM







How can I increase/decrease (frequency/pitch) and phase using fft/ifft



 I think I have the basic code but I’m not sure what to do next







PS: Thanks for the help on the last question everyone I decided not to use the 
FOR loop and sin/cos values and just use fft/ifft











to see if this will work.



Example I have a signal that repeats 1 time every second and I want to



 have it repeat 3 times a second instead.



  %Voiceprint raise lower freq phase conjugate signal



 tic











 clear all, clc,clf,tic



 %% Sound /beep calculation complete



 filerawbeepStr='calculations_complete.wav';



 filerawbeeppathStr='/home/rat/Documents/octave/raw/';



 filevoiceprepathStr='/home/rat/Documents/octave/eq_research/main/











 transform/voice/';



 filewavpathStr='/home/rat/Documents/octave/eq_research/main/transform/



 wav/';



 [ybeep, Fsbeep, nbitsbeep] =



 wavread(strcat(filerawbeeppathStr,filerawbeepStr));



 %addpath(”/home/rat/Documents/octave/eq_research/main/transform/”);











 %add path to location of functions



 %1a voice print import



 [vp_sig_orig, fs_rate, nbitsraw] =



 wavread(strcat(filevoiceprepathStr,'voice8000fs.wav'));



 %vp_sig_orig=vp_sig_orig’;











 vp_sig_len=length(vp_sig_orig);



 %2a create frequency domain



 ya_fft = fft(vp_sig_orig);



 vp_sig_phase_orig = unwrap(angle(ya_fft));



 %get Magnitude



 ya_fft_mag = abs(ya_fft);











%3a frequency back to time domain



 ya_ifft=real(ifft(ya_fft));



 %adjust frequency/phase here? How?



 vp_sig_new=real(ifft(ya_fft_mag.*exp(i*vp_sig_phase_orig)));



 subplot(3,1,1), plot(vp_sig_orig),title('1 original time domain')











 subplot(3,1,2), plot(ya_ifft),title('2 rebuild time domain')



 subplot(3,1,3), plot(vp_sig_new),title('3 adjusted time')



















-----Inline Attachment Follows-----







_______________________________________________



Help-octave mailing list



address@hidden



https://mailman.cae.wisc.edu/listinfo/help-octave







So, regarding your











"



Example I have a signal that repeats 1 time every second and I want to



 have it repeat 3 times a second instead.



"







- why do you need FFT in the first place ?







I.e. if 'x' is your signal, why not simply write







x = linspace(0, 1, 10); % or whatever other way to generate your signal



y = [x x x]; % repeat x 3 times



plot(y);











?







Regards,



  Sergei.









--




Sorry, but this is mostly nonsense. FFT can't be faster than just moving

data in memory - the latter is my solution.



If your audio comes in frequency domain, then by just _one_ 'ifft' you

convert it into time domain, and then my solution works.



You have already been given a link to 'resample' function.



You appear not to understand fundamental things regarding pitch shift. If

your signal gets repeated a number of times, it is not pitch shift.



Pitch shift does not imply signal repetitions and does not imply change

of number of output samples.



AFAIK pitch shift is implemented through overlapping relatively (compared

to the length of the whole musical piece) short FFTs, and the spectrum is

shifted (rather, scaled - you typically need all spectral componenets to be 
multiplied by the same factor) in order to achieve pitch shift - number of 
samples, as I said, does _not_ change.



Start from http://en.wikipedia.org/wiki/Audio_timescale-pitch_modification -> 
http://en.wikipedia.org/wiki/Audio_timescale-pitch_modification#Pitch_scaling .





Regards,

  Sergei.






-----Inline Attachment Follows-----

_______________________________________________
Help-octave mailing list
address@hidden
https://mailman.cae.wisc.edu/listinfo/help-octave

I do not know the exact cause, but this:

t=0:Ts:1-Ts; %sampling period

line is silently wrong. Start from _carefully_ reading

http://en.wikipedia.org/wiki/Floating_point
http://en.wikipedia.org/wiki/Radix
.

Pay attention to:

1) periodic vs non-periodic fractions ->
2) inevitable loss of precision due to periodic fractions anywhere in the
chain of your calculations.

Bear in mind that in modern computers working in _binary_ radix many
"typical" decimal fractions which are non-periodic in decimal become
periodic in binary.

...

I do not know why you want to implement pitch shift yourself. There are
already a number of FOSS implementations. For example, visit

sox.sf.net

, read the manual, pay special attention to 'bend' under

Supported Effects
.

SoX is a _very_ powerful tool, but requires careful reading of its
manpages.

IIRC, 'alsaplayer' has pitch shifting too. Maybe also a 'gstreamer'
plugin.

...

FWIW, 'octave' has 'wavwrite' and 'wavread' functions, and SoX, of course,
supports WAV files (as well as other already mentioned candidates). So
if you need pitch shift as a piece of otherwise 'octave'-based flow, there
is no problem using an external tool doing the pitch shift part of the
flow.

Regards,
  Sergei.