harmonicmodel.m
Go to the documentation of this file.
1 function y = harmonicmodel(x, fs, w, N, t, nH, minf0, maxf0, f0et, maxhd)
2 % Analysis/synthesis of a sound using the sinusoidal harmonic model
3 % x: input sound, fs: sampling rate, w: analysis window (odd size),
4 % N: FFT size (minimum 512), t: threshold in negative dB,
5 % nH: maximum number of harmonics, minf0: minimum f0 frequency in Hz,
6 % maxf0: maximim f0 frequency in Hz,
7 % f0et: error threshold in the f0 detection (ex: 5), % maxhd: max. relative deviation in harmonic detection (ex: .2)
8 % y: output sound
9 M = length(w); % analysis window size
10 Ns= 1024; % FFT size for synthesis
11 H = 256; % hop size for analysis and synthesis
12 N2 = N/2+1; % size postive spectrum
14 hNs = Ns/2; % half synthesis window size
15 hM = (M-1)/2; % half analysis window size
16 pin = max(hNs+1,1+hM); % initialize sound pointer to middle of analysis window
18 fftbuffer = zeros(N,1); % initialize buffer for FFT
19 y = zeros(soundlength+Ns/2,1); % output sound
20 w = w/sum(w); % normalize analysis window
21 sw = zeros(Ns,1);
22 ow = triang(2*H-1); % overlapping window
23 ovidx = Ns/2+1-H+1:Ns/2+H; % overlap indexes
24 sw(ovidx) = ow(1:2*H-1);
25 bh = blackmanharris(Ns); % synthesis window
26 bh = bh ./ sum(bh); % normalize synthesis window
27 sw(ovidx) = sw(ovidx) ./ bh(ovidx);
28 while pin<pend
29  %-----analysis-----%
30  xw = x(pin-hM:pin+hM).*w(1:M); % window the input sound
31  fftbuffer(:) = 0; % reset buffer
32  fftbuffer(1:(M+1)/2) = xw((M+1)/2:M); % zero-phase window in fftbuffer
33  fftbuffer(N-(M-1)/2+1:N) = xw(1:(M-1)/2);
34  X = fft(fftbuffer); % compute the FFT
35  mX = 20*log10(abs(X(1:N2))); % magnitude spectrum
36  pX = unwrap(angle(X(1:N/2+1))); % unwrapped phase spectrum
37  ploc = 1 + find((mX(2:N2-1)>t) .* (mX(2:N2-1)>mX(3:N2)) ...
38  .* (mX(2:N2-1)>mX(1:N2-2))); % find peaks
39  [ploc,pmag,pphase] = peakinterp(mX,pX,ploc); % refine peak values
40  f0 = f0detection(mX,fs,ploc,pmag,f0et,minf0,maxf0); % find f0
41  hloc = zeros(nH,1); % initialize harmonic locations
42  hmag = zeros(nH,1)-100; % initialize harmonic magnitudes
43  hphase = zeros(nH,1); % initialize harmonic phases
44  hf = (f0>0).*(f0.*(1:nH)); % initialize harmonic frequencies
45  hi = 1; % initialize harmonic index
46  npeaks = length(ploc); % number of peaks found
47  while (f0>0 && hi<=nH && hf(hi)<fs/2) % find harmonic peaks
48  [dev,pei] = min(abs((ploc(1:npeaks)-1)/N*fs-hf(hi))); % closest peak
49  if ((hi==1 || ~any(hloc(1:hi-1)==ploc(pei))) && dev<maxhd*hf(hi))
50  hloc(hi) = ploc(pei); % harmonic locations
51  hmag(hi) = pmag(pei); % harmonic magnitudes
52  hphase(hi) = pphase(pei); % harmonic phases
53  end
54  hi = hi+1; %increase harmonic index
55  end
56  hloc(1:hi-1) = (hloc(1:hi-1)~=0).*((hloc(1:hi-1)-1)*Ns/N+1); % synth. locs
57  %-----synthesis-----%
58  Yh = genspecsines(hloc(1:hi-1),hmag,hphase,Ns); % generate sines
59  yh = fftshift(real(ifft(Yh))); % sines in time domain
60  y(pin-hNs:pin+hNs-1) = y(pin-hNs:pin+hNs-1) + sw.*yh(1:Ns); % overlap-add
61  pin = pin+H; % advance the input sound pointer
62 end
Definition: start.py:1
magnitude spectrum pX
Definition: stft_peak.m:24
FFT size for synthesis H
Definition: harmonicmodel.m:11
half analysis window size pin
Definition: harmonicmodel.m:16
function minf0
Definition: harmonicmodel.m:1
function f0et
Definition: harmonicmodel.m:1
if max(w)>1 w=0.9 *w/max(w)
About Git write you should know how to use GIT properly Luckily Git comes with excellent documentation git help man git shows you the available git< command > help man git< command > shows information about the subcommand< command > The most comprehensive manual is the website Git Reference visit they are quite exhaustive You do not need a special username or password All you need is to provide a ssh public key to the Git server admin What follows now is a basic introduction to Git and some FFmpeg specific guidelines Read it at least if you are granted commit privileges to the FFmpeg project you are expected to be familiar with these rules I if not You can get git from etc no matter how small Every one of them has been saved from looking like a fool by this many times It s very easy for stray debug output or cosmetic modifications to slip in
Definition: git-howto.txt:5
N, 1 zeros()
function ploc
FFT size for synthesis(even) H
#define sample
function maxhd
Definition: harmonicmodel.m:1
Inicial output npeaks
Definition: stpt.m:6
FFT of current buffer mX
Definition: stft_peak.m:23
About Git write you should know how to use GIT properly Luckily Git comes with excellent documentation git help man git shows you the available git< command > help man git< command > shows information about the subcommand< command > The most comprehensive manual is the website Git Reference visit they are quite exhaustive You do not need a special username or password All you need is to provide a ssh public key to the Git server admin What follows now is a basic introduction to Git and some FFmpeg specific guidelines Read it at least if you are granted commit privileges to the FFmpeg project you are expected to be familiar with these rules I if not You can get git from etc no matter how small Every one of them has been saved from looking like a fool by this many times It s very easy for stray debug output or cosmetic modifications to slip please avoid problems through this extra level of scrutiny For cosmetics only commits you should e g by running git config global user name My Name git config global user email my email which is either set in your personal configuration file through git config core editor or set by one of the following environment VISUAL or EDITOR Log messages should be concise but descriptive Explain why you made a what you did will be obvious from the changes themselves most of the time Saying just bug fix or is bad Remember that people of varying skill levels look at and educate themselves while reading through your code Don t include filenames in log Git provides that information Possibly make the commit message have a descriptive first an empty line and then a full description The first line will be used to name the patch by git format patch Renaming moving copying files or contents of making those normal commits mv cp path file otherpath otherfile git add[-A] git commit Do not rename or copy files of which you are not the maintainer without discussing it on the mailing list first Reverting broken commits git revert< commit > git revert will generate a revert commit This will not make the faulty commit disappear from the history git reset< commit > git reset will uncommit the changes till< commit > rewriting the current branch history git commit amend allows to amend the last commit details quickly git rebase i origin master will replay local commits over the main repository allowing to merge or remove some of them in the process Note that the reset
Definition: git-howto.txt:153
function w
Definition: harmonicmodel.m:1
#define M(a, b)
Definition: vp3dsp.c:43
pphase
Definition: stft_peak.m:27
analysis window size Ns
Definition: harmonicmodel.m:10
end end
function magnitudes and phases
Plot spectral magnitude
function f0
return end harmonic
Definition: extra/TWM.m:29
frame
Definition: stft.m:14
Discrete Time axis x
function y
Definition: harmonicmodel.m:1
function nH
Definition: harmonicmodel.m:1
length of input sound array hNs
Definition: harmonicmodel.m:14
#define zero
Definition: regdef.h:64
these buffered frames must be flushed immediately if a new input produces new the filter must not call request_frame to get more It must just process the frame or queue it The task of requesting more frames is left to the filter s request_frame method or the application If a filter has several the filter must be ready for frames arriving randomly on any input any filter with several inputs will most likely require some kind of queuing mechanism It is perfectly acceptable to have a limited queue and to drop frames when the inputs are too unbalanced request_frame This method is called when a frame is wanted on an output For an input
ow
Definition: harmonicmodel.m:22
phase spectrum(unwrapped) ploc
overlapping window(triangular window to avoid too much overlapping) ovidx
overlapping window ovidx
Definition: harmonicmodel.m:23
int size
clear max peak[Mmag2, Mloc2]
Definition: extra/TWM.m:16
use a maximum of peaks[f0, f0error]
Sampled sinusoid X
bh
Definition: harmonicmodel.m:25
last sample to start a frame fftbuffer
Definition: harmonicmodel.m:18
initialize buffer for FFT yh
size postive spectrum soundlength
Definition: harmonicmodel.m:13
normalize analysis window sw
Definition: harmonicmodel.m:21
half synthesis window size hM
Definition: harmonicmodel.m:15
sound(x3, Fs)
function fs
Definition: harmonicmodel.m:1
int index
Definition: gxfenc.c:89
function pmag
fftbuffer, N fft()
FFmpeg Automated Testing Environment ************************************Table of Contents *****************FFmpeg Automated Testing Environment Introduction Using FATE from your FFmpeg source directory Submitting the results to the FFmpeg result aggregation server FATE makefile targets and variables Makefile targets Makefile variables Examples Introduction **************FATE is an extended regression suite on the client side and a means for results aggregation and presentation on the server side The first part of this document explains how you can use FATE from your FFmpeg source directory to test your ffmpeg binary The second part describes how you can run FATE to submit the results to FFmpeg s FATE server In any way you can have a look at the publicly viewable FATE results by visiting this as it can be seen if some test on some platform broke with their recent contribution This usually happens on the platforms the developers could not test on The second part of this document describes how you can run FATE to submit your results to FFmpeg s FATE server If you want to submit your results be sure to check that your combination of OS and compiler is not already listed on the above mentioned website In the third part you can find a comprehensive listing of FATE makefile targets and variables Using FATE from your FFmpeg source directory **********************************************If you want to run FATE on your machine you need to have the samples in place You can get the samples via the build target fate rsync Use this command from the top level source this will cause FATE to fail NOTE To use a custom wrapper to run the pass target exec to configure or set the TARGET_EXEC Make variable Submitting the results to the FFmpeg result aggregation server ****************************************************************To submit your results to the server you should run fate through the shell script tests fate sh from the FFmpeg sources This script needs to be invoked with a configuration file as its first argument tests fate sh path to fate_config A configuration file template with comments describing the individual configuration variables can be found at doc fate_config sh template Create a configuration that suits your based on the configuration template The slot configuration variable can be any string that is not yet but it is suggested that you name it adhering to the following pattern< arch >< os >< compiler >< compiler version > The configuration file itself will be sourced in a shell therefore all shell features may be used This enables you to setup the environment as you need it for your build For your first test runs the fate_recv variable should be empty or commented out This will run everything as normal except that it will omit the submission of the results to the server The following files should be present in $workdir as specified in the configuration it may help to try out the ssh command with one or more v options You should get detailed output concerning your SSH configuration and the authentication process The only thing left is to automate the execution of the fate sh script and the synchronisation of the samples directory FATE makefile targets and variables *************************************Makefile can be set to
Definition: fate.txt:142
1:W2 xw()
function N
Definition: harmonicmodel.m:1
initialize sound pointer to middle of analysis window pend
Definition: harmonicmodel.m:17
the buffer and buffer reference mechanism is intended to as much as expensive copies of that data while still allowing the filters to produce correct results The data is stored in buffers represented by AVFilterBuffer structures They must not be accessed but through references stored in AVFilterBufferRef structures Several references can point to the same buffer
these buffered frames must be flushed immediately if a new input produces new output(Example:frame rate-doubling filter:filter_frame must(1) flush the second copy of the previous frame, if it is still there,(2) push the first copy of the incoming frame,(3) keep the second copy for later.) If the input frame is not enough to produce output
function t
Definition: harmonicmodel.m:1
function maxf0
Definition: harmonicmodel.m:1
const char int length
Definition: avisynth_c.h:668
float min
hop size for analysis and synthesis N2
Definition: harmonicmodel.m:12
these buffered frames must be flushed immediately if a new input produces new the filter must not call request_frame to get more It must just process the frame or queue it The task of requesting more frames is left to the filter s request_frame method or the application If a filter has several the filter must be ready for frames arriving randomly on any input any filter with several inputs will most likely require some kind of queuing mechanism It is perfectly acceptable to have a limited queue and to drop frames when the inputs are too unbalanced request_frame This method is called when a frame is wanted on an output For an it should directly call filter_frame on the corresponding output For a if there are queued frames already one of these frames should be pushed If the filter should request a frame on one of its repeatedly until at least one frame has been pushed Return values