Version 1.4.4 25/2 2014 * Mostly bug fixes and doc updates. * Reworked build system on Linux and Mac OS. * Most of the code moved from Mex/Oct interfaces to the backend. * The C code was revised: consistence of argument naming and order, introduced ltfatInt data type for array lengths. * New block processing demos: real-time CQT and Erblets (fast enough only in Matlab). * Added routines for calculating Gabor dual frames using convex optimization (requires UNLocBoX http://unlocbox.sourceforge.net/). * Improved usability of the blockproc GUI: Added possibility to set initial window positions, capturing Ctrl-C shortcut. * Closed #22: Wilson and MDCT transforms are now completely implemented in C. * Renamed framematrix to frsynmatrix Version 1.4.3 22/11 2013 * Fixed bug in nsgabframebounds.m * Added chirped Z-transform * Block processing demos now work in Octave (on Linux) * Backend library now uses C99 complex number format only * Faster block processing via block_interface MEX function * New zoomig features for blockplot Version 1.4.2 17/9 2013 * Added mexExecuter to speed up cell array handling in the backends * All filterbank algorithms are now in the backend * franalasso now packs the FISTA algorithm * More block processing demos: pitch shifting * Added the generalized Goertzel algorithm Version 1.4.1 18/7 2013 * Major change in the output format from the wfilt_ functions. If a wfilt_ function generates a tigth frames, its two outputs are now identical. * Close #67, #68, Mex compilation now works on Mac * Experimental filter backend added to handle filters defined on the frequency side and on the time side using structs and anonymous functions * Limited support for fractional downsampling in filters and filterbanks * erbfilters routine added to generate Erb-spaced filterbanks * Fixed bug #6, demo_audioshrink now works again * All DCT and DST routines now call FFTW directly. * Fixed bug #55, FWT on Octave can now handle complex values * Added floor23, floor234, ceil23 and ceil235 to find the next nice number. Useful for constructing downsampling rates in filterbanks. Version 1.4.0 3/6 2013 * All routines calling the C backend now support single precision * Frames framework has been rewritten for greater speed using anonymous functions * New "operators" directory for the comming inclusion of more operator classes * First alpha version of the block processing framework introduced in "blocks" * The noshearlength routine computes next efficient transform length for a Gabor system on a non-separable lattice * The frames framework now support non-stationary Gabor systems * Compilation of functions calling BLAS and LAPACK has been fixed, so gabdual and gabtight now works in C on all platforms * Better speed when computing many Hermite functions, support for orthonormalization in the sampled continous case * Fast and discrete fractional Fourier transform added * cqt and erblett transforms added Version 1.3.1 5/4 2013 * Fixed compilation on Unix * Wavelets now works in Octave * Improved firwin featuring all the windows from the Nuttall paper Version 1.3.0 20/3 2013 * This is the first full release of the wavelets. Too many changes to list here, but the major features are listed below: * fwt - Fast wavelet transform * fwt2 - 2D FWT with different layouts * ufwt - Undecimated FWT * wfbt - Wavelet filter bank tree * wpfbt - Wavelet packet filter bank tree * wpbest - Best basis search of a wavelet packet tree * wfilt_ functions defines a lot of different possible wavelet and scaling functions. * Mex compiled code is now supported for Windows 64 bit (Windows Vista, 7). Support for Windows 32 has been dropped. * Color test image, lichenstein, added and jpeg color model support in rgb2jpeg * frame multipliers added with the usual functions, framemul, framemulinv, framemuladj, framemuleigs and framemulappr Version 1.2.0 13/12 2012 * Full support for non-separable Gabor lattices with support in the C backend. * Improved non-stationary Gabor systems: bugfixes for system with odd-length shifts, tester has been extented to cover all these cases. * Iterative analysis and synthesis for frames: franaiter and frsyniter uses the conjugate gradients algorithm. * The "frames" framework has changed so that each frame object only includes one frame. This means that you will need to create two frames if you want to perform analysis/synthesis with a bi-orthogonal / canonical dual system. On the other hand, a lot of duplication was removed. * Small bugfixes: idgt2, gabdualnorm Version 1.1.2 2/10 2012 * Almost full support for non-separable Gabor laticces * Multi-win support re-enabled in gabdual and gabtight * Demos finally converted to new documentation system Version 1.1.1 30/3 2012 * Initial inclusion of the frames framework * New and more flexible groupthresh * Much improved framelasso and framegrouplasso replaces the old lasso methods Version 1.0.0 16/6 2011 * Auditory scales: Erb, bark, mel * Gammatone filters. * Filterbanks with a full set of support functions. * non-stationary Gabor frames with a full set of support functions. * rangecompress and rangeexpand does ulaw and alaw. * cocktailparty test signal replaces older 'greasylong' * plot functions for visualizing coefficients of all transforms. * C implementation improved: speedup in gabdual and gabtight, implementation of dgtreal, pfilt and ufilterbank. * nextfastfft computes next larger problem size with a fast FFT. * isgramreal can use BFGS method, requires external software. Version 0.98.2 25/3 2011 * Added C code for IDGT using FIR filters. * WinXP compilation now works without LCC. Version 0.98.1 25/2 2011 * New iterative spectrogram reconstruction featuring the word "LTFAT". * Features added to ltfatarghelper to support importing definitions from aux. functions. Version 0.98 28/1 2011 * The flags 'freqinv' and 'timeinv' can be passed to the DGT, IDGT, DGTREAL and IDGTREAL to select a time- or frequency-invariant phase. * Three new functions to ramp a signal (create a smooth transition from 0 to 1), RAMPUP, RAMPDOWN and RAMPSIGNAL. * nuttall window added to FIRWIN. General cleanup of FIRWIN. If is now possible to taper the window in the middle. * Support for different normalization of the function in all window functions. This is done through the function NORMALIZE. * PGAUSS takes options for shifting the center frequency and specifying the bandwidth, in both samples or Hz. * PINKNOISE: Pink noise generator. * ISGRAM: Spectrogram reconstruction using the classical iterative method by Griffin and Lim. * ELITISTHRESH: Elitist LASSO thresholding. * PRECT and PSINC: periodic rectangular and periodic Sinc function. Version 0.97.2: * The GPC source code is now distributed with LTFAT. A popup dialog has been added to mulaclab to explan the license conditions. * The algorithm for computing dgtreal with a FIR window is now implemented in C. Version 0.97.1: * Support for Octave on Windows XP. * It is now possible to specify various targets and commands in ltfatmex. Version 0.97 * Toolbox is now built upon a standalone C library. * The 'mulaclab' is a graphical user interface for manipulating the spectrogram of a signal. The gui works only in Matlab. * All functions in the LTFAT C library are now available in both single and double precision * Compilation and interfaces for both Matlab and Octave interfaces now works on Windows XP. * It is now possible to supply a window described by a text string or a cell array to all relevant functions. See the help on gabwin or wilwin for a description of the possibilities. * Much better support for optional arguments in functions, and for setting default at startup. See the function ltfatsetdefaults, ltfatgetdefaults and ltfatarghelper * GABRIESZBOUNDS: compute Gabor Riesz bounds for a Gabor Riesz sequence. * WIL2RECT and RECT2WIL: arrange Wilson coefficients in a rectangular shape (with holes) at the correct position in the TF-plane. * PEVEN and PODD extracts the even and odd part of a signal. Version 0.96 12/1 2009 svn no 728 * Matlab MEX compilation now works under Windows. See the instructions in the INSTALL file. * Speed optimizations in the C-code used by DGT, DWILT and MDCT and their inverses. * New functions DGTREAL and IDGTREAL works with the positive frequencies of the DGT of real valued signals. * New functions FFTREAL computes only the positive frequencies of the FFT of a real valued input signal. * More systematic naming of functions: CANDUAL -> GABDUAL CANTIGHT -> GABTIGHT MIXDUAL -> GAMIXDUAL PROJDUAL -> GABPROJDUAL GFBOUNDS -> GABFRAMEBOUNDS and GABRIESZBOUNDS TF_ADAPTLASSO -> GABELITISTLASSO TF_GROUPLASSO -> GABGROUPLASSO * Reassignment is a method for sharpening the spectrogram. Support for reassignment is included in the new function REASSIGN and an easy to use plot RESGRAM. * Easy to use plot for plotting instantantaneous frequency: INSTFREQPLOT * Three different methos for computing instantaneous time and frequency: INSTTFDGT, INSTTFPHASE and INSTTFABS. * General speedup of many of the SPREAD* routines based on speedup in COL2DIAG and more efficient algorithms for sparse matrices. * COL2DIAG provides the basic coordinate change needed for efficient implementation of spreading function methods. COL2DIAG has a C-implementation. * New function WIL2RECT converts Wilson coefficients from the standard compact layout to a more loose layout, where the coefficients are appropriatly placed on the TF-plane. The rectangular format is welll suited for visualizing Wilson coefficients. * The functionality of GFBOUNDS was split into two methods computing either frame bounds or Riesz basis bounds * Dynamic range in SGRAM and RESGRAM is now specified by the 'dynrange' parameter instead of previously 'range'. * greasylong and doppler signals added. * Periodic Heaviside function added, PHEAVISIDE. * Simple exponential wave added as EXPMODE. Version 0.95 6/3 2008 svn no. 595 * DCT based resampling function. Version 0.94 24/10 2007 svn no. 556 * Numerically stable computation of Hermite functions. Thanks to Thomasz Hrycak. * gabmulappr (approximation of an operator by a Gabor multiplier) now works with fast algorithm. * group lasso shrinkage and adaptive lasso shrinkage added with an example (examp_audioshrink) * Removed all support of lattices in the spreading operator routines, as this is not practically usefull. * Special support in candual for windows shorter than the number of channels. * The configure style system has been removed. Use ltfatmex instead. * phaseplot now uses the phaselocked dgt by default. Version 0.93 10/8 2007 svn no. 504 * Easy compilation of Mex/Octave interfaces by 'ltfatmex' command * Bug fixed for Wilson bases. * Better support of choosing an alternative dimension for the various transforms. * fmax option added to sgram * fftresample does Fourier interpolation * phaseplot changed to always do full STFT * moved to GPL v 3.0 license