CHANGELOG

Changes in Rubber Band v4.0

This release adds a new API that is simpler than the existing one in
cases where only pitch-shifting is required.

 * Add a new API, RubberBandLiveShifter, which is simpler to use
   than the general RubberBandStretcher interface in cases where only
   pitch-shifting is required. For more general purposes the original
   interface is still the proper one.
 * Fix incorrect URIs in LV2 plugins
 * Fix stack overflow in seldom used R2 time-domain smoothing code
 * Add support for key-frame maps (and the live shifter) to JNI interface
 * Fix some compatibility issues and warnings with certain compilers

Apart from this entirely new class, the rest of the API is unchanged
and the library is binary compatible with existing code back to
version 1.7.


Changes in Rubber Band v3.3

This is a bug-fix release with no changes to audio quality.

 * Fix incorrect output with unexpectedly large buffers in process
 * Add getProcessSizeLimit to query the maximum buffer length for
   process and highest supported value for setMaxProcessSize
 * Fix formal test failure in some contexts

One new function has been added to the API. The library is otherwise
binary compatible with existing code back to version 1.7.


Changes in Rubber Band v3.2.1

This is a build-fix release with no new features.

 * Fix formal test failure in build configurations using other
   resamplers than the built-in one

The API is unchanged from 3.0 and the library is binary compatible
back to version 1.7.


Changes in Rubber Band v3.2

This significant release contains a number of improvements, including
some output quality enhancements for certain processing modes.

 * Fix inability of the R3 engine to produce mono-compatible output
   when processing stereo. In R2 the OptionChannelsTogether option has
   always done this (with some quality caveats). In R3 this option was
   more nearly mono-compatible than the default, but not
   entirely. With this release, OptionChannelsTogether uses a new
   method that is both high quality and mono-compatible and is
   recommended for situations benefiting from this.
 * Correct the start delay calculation in "resample-before" modes of
   both stretcher engines. This affects latency reporting in realtime
   mode when using the non-recommended realtime pitch options
   (i.e. those other than OptionPitchHighConsistency, which is the
   recommended option for variable-rate use in realtime mode)
 * Relatedly, fix failure to use OptionPitchHighConsistency in R3
   LADSPA/LV2 plugins
 * Fix small difference in output from R3 engine depending on whether
   time and pitch ratios were provided on construction or immediately
   after it
 * Expand automated tests

The API is unchanged from 3.0 and the library is binary compatible
back to version 1.7.


Changes in Rubber Band v3.1.3

This is a bug-fix release with no new features or changes to audio
quality.

 * Fix failure to generate output when called with final=true and no
   samples at input (a regression in v3.1, this worked correctly in
   v3.0 and in the R2 engine)
 * Fix occasional truncation of final block when using Speex resampler
 * Fix hang in the command-line utility code at end of processing with
   certain file formats

The API is unchanged from 3.0 and the library is binary compatible
back to version 1.7.


Changes in Rubber Band v3.1.2

This minor release contains a small number of build fixes.

 * Make all optional targets manually toggleable, using Meson
   "features" options (see COMPILING.md for documentation)
 * Further fix to JNI autodetection, following that of 3.1.1 -
   hopefully solving the problem completely this time!
 * Correct the source file listing in the Android makefile
 * Fix build with NO_EXCEPTIONS, e.g. for Android
 * Fix required #includes for mingw32 builds

The API is unchanged from 3.0 and the library is binary compatible
back to version 1.7.


Changes in Rubber Band v3.1.1

This minor release contains no code changes, only a fix to the build
system.

 * Fix failure to configure and build on some systems without a Java
   compiler. Java/JNI support is entirely optional in Rubber Band, but
   auto-detection of whether to build it or not was failing in certain
   environments: this broke the normal build process for some users.

The API is unchanged from 3.0 and the library is binary compatible
back to version 1.7.


Changes in Rubber Band v3.1

This significant release contains many enhancements that can improve
performance, as well as a number of bug fixes.

 * Add support for "short window" mode, OptionWindowShort, to the new
   R3 processing engine. This option previously only took effect with
   the R2 engine. When used with R3 it disables the R3 multi-window
   logic while retaining R3's signal analysis, yielding output with
   some of the quality advantages of R3 (especially for non-percussive
   content) but with far lower CPU usage and processing delay.
 * Implement OptionPitchHighSpeed in the R3 engine. Previously it
   always used OptionPitchHighConsistency internally. With this option
   real-time pitch-shifting upwards becomes much faster, at the
   expense of some output quality. (However OptionPitchHighConsistency
   is still always necessary to avoid artifacts when performing
   dynamic pitch changes.)
 * Add support for the SLEEF library (libsleefdft) for FFTs. In
   our tests this library approaches the performance of vDSP and IPP
   and improves on the other supported options.
 * Add support for external Speex library (libspeexdsp) for
   resampling. This is a very fast resampler, significantly improved
   over the older Speex code that has historically been bundled with
   Rubber Band. It may still introduce a little more audible noise on
   ratio changes than the built-in resampler.
 * Update the R2 engine so that it always restores vertical phase
   coherence gradually when returning to a ratio of 1.0 (as R3 also
   does) rather than waiting for silence or a transient.
 * Update R3 to properly support all sample rates between 8000 and
   192000 Hz.
 * Update command-line utility so as to use the extension of the
   output filename to determine the file type to write. Previously
   the output was always the same format as the input file,
   regardless of extension.
 * Remove support for libresample. This integration has not been
   exercised in many years and other resamplers are better options.
 * Fix several bugs, including a crash when run with very low sample
   rates.

The API is unchanged from 3.0 and the library is binary compatible
back to version 1.7.


Changes in Rubber Band v3.0.0

 * Introduce a new processing engine, the R3 (Finer) engine, which
   typically produces higher-quality output than the existing R2
   engine, though at significantly higher CPU cost. The R2 engine is
   still the default, and R3 can be selected using the new
   OptionEngineFiner option on construction. See the documentation for
   more details.
 * Add ability to provide a custom set of log callbacks, so that
   debug and warning logs can be routed to the application's log
   stream and/or handled in a realtime-safe way
 * Add option to shift formant independently of pitch (R3 engine
   only)
 * Deprecate the rather ambiguous function getLatency(), effectively
   replacing it with two new functions getPreferredStartPad() and
   getStartDelay(). See their documentation for more details

The library is both binary and API compatible all the way back to the
1.x series for existing applications.  Code written to use 3.0 is not
necessarily compatible with 2.x or 1.x, as several new functions and
enum values have been added, but code written to use any earlier
version can update to 3.0 without modification (and will continue to
use the same processing engine if the calling code is unchanged).


Changes in Rubber Band v2.0.2

 * Fix a crash in certain configurations when using mid-side
   processing (OptionChannelsTogether) with pitch shifts
 * Fix failure to build on very old macOS versions (<= 10.8) when
   combined with much newer compilers

The API is unchanged and the library is binary compatible back to
version 1.7.


Changes in Rubber Band v2.0.1

 * Add a preliminary LV2 version of the example pitch-shifter plugin,
   and update the existing LADSPA version to fix failure of the wet/dry
   control to work correctly under in-place use
 * Make the built-in resampler (introduced in v2.0.0) the default for
   all builds, and use it in the official packaged versions of the
   command-line utility. External resampler libraries are still
   available as build options
 * A couple of build fixes, including supporting the proper Meson
   option (default_library=shared/static/both) at configuration time
   to choose which type of library to build, instead of the custom
   option used since v1.9.1

The API is unchanged and the library is binary compatible back to
version 1.7.


Changes in Rubber Band v2.0.0

 * Introduce a built-in resampler, intended to minimise artifacts on
   pitch changes and to allow Rubber Band to be compiled without any
   external dependencies (when also using the built-in FFT). The
   built-in resampler is newer and less well-tested than
   libsamplerate, and is not yet the default except in no-dependency
   builds
 * Introduce single compilation unit single/RubberBandSingle.cpp which
   includes the whole library, to allow adding Rubber Band Library to
   a C++ project without having to build anything separately. This
   version always uses the built-in FFT and resampler and does not
   support any extra configuration
 * Update the stretch calculation logic for real-time mode, making it
   more rigorous about avoiding timing drift when the pitch ratio
   changes frequently
 * Update the example LADSPA pitch shift plugin, adding a wet/dry mix
   control to test (and take advantage of) the better timing
   management in real-time mode
 * Add a pitch/frequency-map file option to the command-line utility,
   allowing time-varying pitch shifts to be pre-programmed

The API is unchanged and the library is still binary compatible back
to version 1.7.

We incremented the major version number, not because of any API
change, but because of a change to the timing of audio output for some
applications of real-time mode. While the library is code-compatible
with earlier releases, it is not "output-compatible" and may change
automation-driven mixes even when used with identical options. If you
are already using an earlier release in real-time mode, please test
v2.0.0 carefully before updating.

Note that there are no pure bug-fixes in v2.0.0: if you are happy with
v1.9.2, there is no pressing reason to update.


Changes in Rubber Band v1.9.2

 * Introduce a built-in FFT option to be a good default when no
   requirements point to a particular external library preference
 * Fix potential crash following some reconfigure options. Thanks
   to Michael Tyson for reporting this
 * Ensure Accelerate is listed in generated pkg-config if required
 * Check for existence of sincos in configuration
 * Fix host identification when cross-compiling

The API is unchanged and the library is binary compatible with
version 1.7.


Changes in Rubber Band v1.9.1

 * Switch build system from Makefiles and Visual Studio project to
   Meson/Ninja for all platforms. There are still Makefiles and VS
   projects included in otherbuilds/ for those who wish to use them to
   build the static library directly
 * Make various fixes to improve sound quality when pitch-shifting
   dynamically in real-time (requires libsamplerate)
 * Fix floating-point exception when a very very long stretch factor
   is presented
 * Move the two directories that together provide the .NET interface
   (rubberband-sharp and rubberband-dll) into their own subdir (dotnet)
 * Ensure the library builds and runs correctly on ARM Mac (Apple
   Silicon, M1), and also on Windows using the Visual C++ Clang
   front-end

The API is unchanged and the library is binary compatible with
version 1.7.


Changes in Rubber Band v1.9

 * Fix incorrect numbering of pitch speed/quality flags in the
   auxiliary C wrapper header. The effect of this was that code using
   the C wrapper that intended to select the higher-quality
   pitch-shift mode was actually choosing the higher-speed mode, and
   vice versa. (The third mode - high-consistency, commonly used in
   real-time applications - was correct.) Thanks to Michael Bradshaw
   for reporting this.
 * Add a .NET interface, generously contributed by Jonathan Gilbert
   under a BSD-like licence
 * Improve error handling in command-line utility
 * Fix failure to build with FFTW_SINGLE_ONLY defined
 * Fix some compiler warnings
 * Install the JNI target (with make install) only if it has actually
   been built, avoiding an error otherwise

The API is unchanged except for the fix noted above, as a result of
which the minor version number has been increased. The library is
binary compatible with version 1.7.


Changes in Rubber Band v1.8.2

 * Fix a number of small memory leaks
 * Make stretcher more robust to being fed invalid input (with NaNs)
 * Include iOS build file
 * Various platform build fixes and improvements

The API is unchanged and the library is binary compatible with
version 1.7.


Changes in Rubber Band v1.8.1

 * Fix a crash in formant-preserving pitch shift for some build targets

The API is unchanged and the library is binary compatible with
version 1.7.


Changes in Rubber Band v1.8

 * Add build support for Win32/MSVC, Android, and various libraries
 * Add Java JNI interface

The API is unchanged and the library is binary compatible with
version 1.7.


Changes in Rubber Band v1.7

 * Add the centre-focus option as an alternative processing mode for
   stereo (using mid-side arrangement)
 * Several bug fixes

The library is binary compatible with version 1.6 for forward
compatibility (values have been added to an existing enum).  Code
written to use 1.7 is not necessarily compatible with 1.6.


Changes in Rubber Band v1.6

 * Add Smoothing option.  This uses a window-presum FFT, introducing
   time-domain aliasing which is then smoothed using a sinc window.
   This can be used in combination with any of the existing processing
   control options.  This will soften transients but the result may
   still be more pleasant for some material that is not very amenable
   to being time stretched.
 * Fix silent channel of output when processing with band-limited
   transients option
 * Include libresample support

The library is binary compatible with version 1.5 for forward
compatibility (values have been added to an existing enum).  Code
written to use 1.6 is not necessarily compatible with 1.5.


Changes in Rubber Band v1.5

 * Add a more reliable transient detection mode, and make the mode
   selectable using OptionDetectorXXX flags -- the new method is
   the default
 * Band-limit transient detectors to avoid being distracted by
   inaudible garbage
 * Add a key-frame mapping facility for variable stretch ratio
   management during offline stretches

The library is binary compatible with version 1.4 for forward
compatibility (a function has been added and an enum changed, but no
existing entry points have changed).  Code written to use 1.5 is not
necessarily compatible with 1.4.


Changes in Rubber Band v1.4 

 * Fix a hang when faced with some very peculiar stretch factors
 * Fix some incorrect threading condition usage
 * Some code rearrangement
 * Fix build on Solaris

The library is binary compatible with version 1.3.


Changes in Rubber Band v1.3

 * Fix a bug that may cause incorrect output during the first process
   block of some audio files, when processing in offline mode
 * Fix a small number of build issues and more minor bugs

The library is binary compatible with version 1.2.


Changes in Rubber Band v1.2

 * Add an initial "formant preservation" option when pitch shifting
 * Real-time pitch shifting now uses a faster method by default, with
   less variation in CPU usage
 * The code is more amenable to compiler auto-vectorization (through
   e.g. gcc --ftree-vectorize).