Version 2.1.1 x/6 2015

* New function for computing higher order phase derivatives: gabphasederiv

* New function for adjustable reassignment: gabreassignadjust

* Fixed bug in gabphasegrad which prevented it to be used with any window

other than Gaussian.

* Fixed bug in lconv.

Version 2.1.0 6/5 2015

* New filterbank-generating function: warpedfilters

* New filterbank spectrogram reassignment function: filterbankreassign

* ltfatarghelper is now also implemented as a MEX file.

* The block function can now handle synchronized simultaneous playback and

recording. E.g. {'rec','file.wav'} plays a wav file while recording.

This is different from the option 'playrec', which works the "other way"

i.e. it records and play the recorded.

* Function framebounds(...,'iter') now uses pcg to represent the inverse of

the frame operator when calculating the lowest eigenvalue using eigs.

* Added implementation of some quadratic time-frequency representations:

Rihaczek, Wigner-Ville

* Fixed #112 (Sourceforge): Octave package now compiles on Windows (mxe version)

thanks to John Donoghue.

* Related to #111 (Sourceforge): Wavelet filters were hand-centered around the origin.

* comp_filterbank, comp_ifilterbank and ltfatarghelper can be compiled as MEX files on Octave.

Version 2.0.1 8/10 2014 (released only at Octave-Forge)

* Included test suite.

Version 2.0.0 7/10 2014

* Fixed bug in filterbankdual and filterbankrealdual for the uniform case.

Also the routines are now much faster.

* Added possibility to force treatment of a filterbank as a painless

filterbank using 'forcepainless' flag in filterbank(real)dual routines.

* Fixed bug in comp_ifilterbank MEX causing crashes for mixed type

filterbanks.

* Added implementation of dual-tree wavelet transform, general dual-tree

wavelet filterbanks and *M*-band transforms.

* New dual-tree wavelet filters: wfiltdt_qshift, wfiltdt_oddeven,

wfiltdt_optsym, wfiltdt_dden

* New regular DWT filters: wfilt_symorth, wfilt_symtight

* wfilt_ functions now return filters in a format compatible with

LTFAT filter format (struct with .h and .offset fields).

* Closed #100 FEATURE: Add on-the-fly sample rate conversion to the

blockproc framework. It effectivelly allows recording and playback at

arbitrary sampling rates.

* Added support for cross-compilation (src/Makefile_crossmingw): Producing

Windows binaries on an Unix system using MXE (http://mxe.cc/). Works only

with MEX files for Matlab.

* Blockproc. framework now supports any sampling rate from interval 4-96kHz

trough on-the-fly resampling in the background.

* Fixed compilation issue with gcc 4.7.

Version 1.4.5 7/5 2014

* Phase locking in DGT ('timeinv' flag) moved to the C backend. It

was too slow in Octave.

* Added franabp - solving a basis pursuit problem

argmin ||c||_1 s.t. Fc = f

for an arbitrary frame F.

* The lambda parameter in franabp and franalasso can now be a vector.

=> thresh acceps such lambda too.

* New demos to accompany the LTFAT 2.0 paper: demo_filterbanks,

demo_wavelets, demo_wfbt, demo_bpframemul, demo_phaseret

* New blockproc demo: demo_blockproc_effects

* blockproc framework can be now used offline without a single call to

the playrec MEX.

* Closed #49: blockwrite method. Allows blockwise storage of audio data

to a wav file.

* All wavelet routines were correctly included in the frames framework.

Correct computation of canonical dual and tight frames for undecimated

wavelet filterbanks.

* New wrapper frame types in the frames framework: erbletfb, cqtfb

* Removed demo_blockproc_pitchshift. It is now part of demo_blockproc_effects.

* Removed some esoteric wavelet filters.

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