c = gla(s,g,a,M) c = gla(s,g,a,M,maxit) c = gla(s,g,a,M,maxit,tol) [c,f,relres,iter] = gla(...)
| s | Initial coefficients. |
| g | Analysis Gabor window |
| a | Hop factor |
| M | Number of channels |
| maxit | Maximum number of iterations. |
| tol | relative tolerance |
| c | Coefficients with the reconstructed phase |
| f | Reconstructed signal. |
| relres | Vector of residuals. |
| iter | Number of iterations done. |
gla(s,g,a,M) attempts to find coefficients c from their abs. value:
c = dgtreal(f,g,a,M,'timeinv'); s = abs(c);
using the Griffin-Lim algorithm.
[c,f,relres,iter]=gla(...) additionally returns an array of residuals relres, the number of iterations done iter and the coefficients c with the reconstructed phase. The relationship between f and c is:
f = idgtreal(c,gd,a,M,'timeinv')
where gd is the canonical dual window obtained by gabdual.
| 'input' | Choose the starting phase as the phase of the input s. This is the default |
| 'zero' | Choose a starting phase of zero. |
| 'rand' | Choose a random starting phase. |
| 'coefmod',coefmod | Anonymous function in a form coefmod = @(c) ...; altering coefficients in each iteration after the phase update has been done. This is usefull when e.g. phase of some of the coefficients is known. |
| 'timemod',timemod | Anonymous function in a form timemod = @(f) ...; altering the time-domain signal in each iteration. This is usefull when e.g. the time support of the signal is known. Note that numel(f)= size(s,2)*a. |
| 'gla' | The original Giffin-Lim iteration scheme. This is the default. |
| 'fgla' | A fast Griffin-Lim iteration scheme from Perraudin et. al.. |
| 'alpha',a | Parameter of the Fast Griffin-Lim algorithm. It is ignored if not used together with 'fgla' flag. |
| 'maxit',n | Do at most n iterations. |
| 'tol',t | Stop if relative residual error is less than the specified tolerance. |
| 'Ls',Ls | Crop the reconstructed signal f to length Ls. |
| 'print' | Display the progress. This is disabled by default. |
| 'printstep',p | If 'print' is specified, then print every p'th iteration. Default value is p=10; |
D. Griffin and J. Lim. Signal estimation from modified short-time Fourier transform. Acoustics, Speech and Signal Processing, IEEE Transactions on, 32(2):236--243, Apr 1984. [ DOI ]
N. Perraudin, P. Balazs, and P. Søndergaard. A fast Griffin-Lim algorithm. In Applications of Signal Processing to Audio and Acoustics (WASPAA), IEEE Workshop on, pages 1--4, Oct 2013. [ DOI ]