mirror of https://github.com/PCSX2/pcsx2.git
3c408de5c9
|
||
|---|---|---|
| .. | ||
| soundtouch | ||
| source | ||
| CMakeLists.txt | ||
| COPYING.TXT | ||
| README.html | ||
| SoundTouch.vcxproj | ||
| SoundTouch.vcxproj.filters | ||
README.html
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<title>SoundTouch library README</title>
<meta http-equiv="Content-Language" content="en-us">
<meta name="author" content="Olli Parviainen">
<meta name="description" content="Readme file for SoundTouch audio processing library">
<style>
body {
font-family: Arial, Helvetica;
}
</style>
</head>
<body class="normal">
<hr>
<h1>SoundTouch audio processing library v2.3.1</h1>
<p class="normal">SoundTouch library Copyright © Olli Parviainen 2001-2021</p>
<hr>
<h2>1. Introduction </h2>
<p>SoundTouch is an open-source audio processing library that allows
changing the sound tempo, pitch and playback rate parameters
independently from each other, i.e.:</p>
<ul>
<li> Sound tempo can be increased or decreased while maintaining the
original pitch</li>
<li> Sound pitch can be increased or decreased while maintaining the
original tempo</li>
<li> Change playback rate that affects both tempo and pitch at the
same time</li>
<li> Choose any combination of tempo/pitch/rate</li>
</ul>
<h3>1.1 Contact information </h3>
<p>Author email: oparviai 'at' iki.fi </p>
<p>SoundTouch WWW page: <a href="http://soundtouch.surina.net">http://soundtouch.surina.net</a></p>
<p>SoundTouch git repository: <a
href="https://gitlab.com/soundtouch/soundtouch.git">https://gitlab.com/soundtouch/soundtouch.git</a></p>
<hr>
<h2>2. Compiling SoundTouch</h2>
<p>Before compiling, notice that you can choose the sample data format if it's
desirable to use 16bit integer sample data instead of floating point samples. See
section "sample data format" for more information.</p>
<p>Also notice that SoundTouch can use OpenMP instructions for parallel
computation to accelerate the runtime processing speed in multi-core systems,
however, these improvements need to be separately enabled before compiling. See
OpenMP notes in Chapter 3 below.</p>
<h3>2.1. Building in Microsoft Windows</h3>
<p>Project files for Microsoft Visual C++ are supplied with the source
code package. Go to Microsoft WWW page to download
<a href="http://www.visualstudio.com/en-US/products/visual-studio-express-vs">
Microsoft Visual Studio Express version for free</a>.
</p>
<p>To build the binaries with Visual C++ compiler, either run
"make-win.bat" script, or open the appropriate project files in source
code directories with Visual Studio. The final executable will appear
under the "SoundTouch\bin" directory. If using the Visual Studio IDE
instead of the make-win.bat script, directories bin and lib may need to
be created manually to the SoundTouch package root for the final
executables. The make-win.bat script creates these directories
automatically. </p>
<p><strong>C# example</strong>: The source code package includes also a C# example
application for Windows that shows how to invoke SoundTouch.dll
dynamic-load library for processing mp3 audio.
<p><strong>OpenMP NOTE</strong>: If activating the OpenMP parallel computing in
the compilation, the target program will require additional vcomp dll library to
properly run. In Visual C++ 9.0 these libraries can be found in the following
folders.</p>
<ul>
<li>x86 32bit: C:\Program Files (x86)\Microsoft Visual Studio
9.0\VC\redist\x86\Microsoft.VC90.OPENMP\vcomp90.dll</li>
<li>x64 64bit: C:\Program Files (x86)\Microsoft Visual Studio
9.0\VC\redist\amd64\Microsoft.VC90.OPENMP\vcomp90.dll</li>
</ul>
<p>In other VC++ versions the required library will be expectedly found in similar
"redist" location.</p>
<p>Notice that as minor demonstration of a "dll hell" phenomenon both the 32-bit
and 64-bit version of vcomp90.dll have the same filename but different contents,
thus choose the proper version to allow the program to start.</p>
<h3>2.2. Building in Gnu platforms</h3>
<p>The SoundTouch library compiles in practically any platform
supporting GNU compiler (GCC) tools.
<h4>2.2.1 Compiling with autotools</h4>
<p>To install build prerequisites for 'autotools' tool chain:</p>
<pre> sudo apt-get install automake autoconf libtool build-essential</pre>
<p>To build and install the binaries, run the following commands in
/soundtouch directory:</p>
<table border="0" cellpadding="0" cellspacing="4">
<tbody>
<tr>
<td style="vertical-align: top;">
<pre>./bootstrap -</pre>
</td>
<td style="vertical-align: top;">Creates "configure" file with
local autoconf/automake toolset.<br>
</td>
</tr>
<tr valign="top">
<td>
<pre>./configure -</pre>
</td>
<td>
<p>Configures the SoundTouch package for the local environment.
Notice that "configure" file is not available before running the
"./bootstrap" command as above.<br>
</p>
</td>
</tr>
<tr valign="top">
<td>
<pre>make -</pre>
</td>
<td>
<p>Builds the SoundTouch library & SoundStretch utility. You can
optionally add "-j" switch after "make" to speed up the compilation in
multi-core systems.</p>
</td>
</tr>
<tr valign="top">
<td>
<pre>make install -</pre>
</td>
<td>
<p>Installs the SoundTouch & BPM libraries to <b>/usr/local/lib</b>
and SoundStretch utility to <b>/usr/local/bin</b>. Please notice that
'root' privileges may be required to install the binaries to the
destination locations.</p>
</td>
</tr>
</tbody>
</table>
<b>Compiling portable Shared Library / DLL version</b>
<p> The GNU autotools compilation does not automatically create a shared-library version of
SoundTouch (.so or .dll) that features position-independent code and C-language
api that are more suitable for cross-language development than C++ libraries.</p>
<p> Use script "make-gnu-dll-sh" to build a portable dynamic library version if such is desired.</p>
<h4><b>2.2.2 Compiling with cmake</b></h4>
<p>'cmake' build scripts are provided as an alternative to the autotools toolchain.</p>
<p>To install cmake build prerequisites:</p>
<pre> sudo apt-get install libtool build-essential cmake</pre>
<p>To build:</p>
<pre>
cmake .
make -j
make install</pre>
<p>To compile the additional portable Shared Library / DLL version with the native C-language API:</p>
<pre>
cmake . -DSOUNDTOUCH_DLL=ON
make -j
make install</pre>
<h3>2.3. Building in Android</h3>
<p>Android compilation instructions are within the
source code package, see file "<b>source/Android-lib/README-SoundTouch-Android.html</b>"
in the source code package. </p>
<p>The Android compilation automatically builds separate .so library binaries
for ARM, X86 and MIPS processor architectures. For optimal device support,
include all these .so library binaries into the Android .apk application
package, so the target Android device can automatically choose the proper
library binary version to use.</p>
<p>The <strong>source/Android-lib</strong> folder includes also an Android
example application that processes WAV audio files using SoundTouch library in
Android devices.</p>
<h3>2.4. Building in Mac</h3>
<p>Install autoconf tool as instructed in <a
href="http://macappstore.org/autoconf/">http://macappstore.org/autoconf/</a>, or alternatively the 'cmake' toolchain.</p>
<p>Then, build as described above in section "Building in Gnu platforms".</p>
<hr>
<h2>3. About implementation & Usage tips <h3>3.1. Supported sample data formats</h3>
<p>The sample data format can be chosen between 16bit signed integer
and 32bit floating point values.</p>
</p> The default sample type is 32bit floating point format,
which also provides better sound quality than integer format because
integer algorithms need to scale already intermediate calculation results to
avoid integer overflows. These early integer scalings can slightly degrade
output quality.</p>
<p> In Windows environment, the sample data format is chosen in file
"STTypes.h" by choosing one of the following defines:</p>
<ul>
<li> <span style="font-weight: bold;">#define
SOUNDTOUCH_INTEGER_SAMPLES</span> for 16bit signed integer</li>
<li> <span style="font-weight: bold;">#define </span><span style="font-weight: bold;">SOUNDTOUCH_</span><span
style="font-weight: bold;">FLOAT_SAMPLES</span> for 32bit floating
point</li>
</ul>
<p> In GNU environment, the floating sample format is used by default,
but integer sample format can be chosen by giving the following switch
to the configure script: </p>
<blockquote>
<pre>./configure --enable-integer-samples</pre>
</blockquote>
<p>The sample data can have either single (mono) or double (stereo)
audio channel. Stereo data is interleaved so that every other data
value is for left channel and every second for right channel. Notice
that while it'd be possible in theory to process stereo sound as two
separate mono channels, this isn't recommended because processing the
channels separately would result in losing the phase coherency between
the channels, which consequently would ruin the stereo effect.</p>
<p>Sample rates between 8000-48000H are supported.</p>
<h3>3.2. Processing latency</h3>
<p>The processing and latency constraints of the SoundTouch library are:</p>
<ul>
<li> Input/output processing latency for the SoundTouch processor is
around 100 ms. This is when time-stretching is used. If the rate
transposing effect alone is used, the latency requirement is much
shorter, see section 'About algorithms'.</li>
<li> Processing CD-quality sound (16bit stereo sound with 44100H
sample rate) in real-time or faster is possible starting from
processors equivalent to Intel Pentium 133Mh or better, if using the
"quick" processing algorithm. If not using the "quick" mode or if
floating point sample data are being used, several times more CPU power
is typically required.</li>
</ul>
<h3>3.3. About algorithms</h3>
<p>SoundTouch provides three seemingly independent effects: tempo,
pitch and playback rate control. These three controls are implemented
as combination of two primary effects, <em>sample rate transposing</em>
and <em>time-stretching</em>.</p>
<p><em>Sample rate transposing</em> affects both the audio stream
duration and pitch. It's implemented simply by converting the original
audio sample stream to the desired duration by interpolating from
the original audio samples. In SoundTouch, linear interpolation with
anti-alias filtering is used. Theoretically a higher-order
interpolation provide better result than 1st order linear
interpolation, but in audio application linear interpolation together
with anti-alias filtering performs subjectively about as well as
higher-order filtering would.</p>
<p><em>Time-stretching </em>means changing the audio stream duration
without affecting it's pitch. SoundTouch uses WSOLA-like
time-stretching routines that operate in the time domain. Compared to
sample rate transposing, time-stretching is a much heavier operation
and also requires a longer processing "window" of sound samples used by
the processing algorithm, thus increasing the algorithm input/output
latency. Typical i/o latency for the SoundTouch time-stretch algorithm
is around 100 ms.</p>
<p>Sample rate transposing and time-stretching are then used together
to produce the tempo, pitch and rate controls:</p>
<ul>
<li> <strong>'Tempo'</strong> control is implemented purely by
time-stretching.</li>
<li> <strong>'Rate</strong>' control is implemented purely by sample
rate transposing.</li>
<li> <strong>'Pitch</strong>' control is implemented as a
combination of time-stretching and sample rate transposing. For
example, to increase pitch the audio stream is first time-stretched to
longer duration (without affecting pitch) and then transposed back to
original duration by sample rate transposing, which simultaneously
reduces duration and increases pitch. The result is original duration
but increased pitch.</li>
</ul>
<h3>3.4 Tuning the algorithm parameters</h3>
<p>The time-stretch algorithm has few parameters that can be tuned to
optimize sound quality for certain application. The current default
parameters have been chosen by iterative if-then analysis (read: "trial
and error") to obtain best subjective sound quality in pop/rock music
processing, but in applications processing different kind of sound the
default parameter set may result into a sub-optimal result.</p>
<p>The time-stretch algorithm default parameter values are set by the
following #defines in file "TDStretch.h":</p>
<blockquote>
<pre>#define DEFAULT_SEQUENCE_MS AUTOMATIC<br>#define DEFAULT_SEEKWINDOW_MS AUTOMATIC<br>#define DEFAULT_OVERLAP_MS 8</pre>
</blockquote>
<p>These parameters affect to the time-stretch algorithm as follows:</p>
<ul>
<li> <strong>DEFAULT_SEQUENCE_MS</strong>: This is the default
length of a single processing sequence in milliseconds which determines
the how the original sound is chopped in the time-stretch algorithm.
Larger values mean fewer sequences are used in processing. In principle
a larger value sounds better when slowing down the tempo, but worse
when increasing the tempo and vice versa.<br>
<br>
By default, this setting value is calculated automatically according to
tempo value.<br>
</li>
<li> <strong>DEFAULT_SEEKWINDOW_MS</strong>: The seeking window
default length in milliseconds is for the algorithm that seeks the best
possible overlapping location. This determines from how wide a sample
"window" the algorithm can use to find an optimal mixing location when
the sound sequences are to be linked back together.<br>
<br>
The bigger this window setting is, the higher the possibility to find a
better mixing position becomes, but at the same time large values may
cause a "drifting" sound artifact because neighboring sequences can be
chosen at more uneven intervals. If there's a disturbing artifact that
sounds as if a constant frequency was drifting around, try reducing
this setting.<br>
<br>
By default, this setting value is calculated automatically according to
tempo value.<br>
</li>
<li> <strong>DEFAULT_OVERLAP_MS</strong>: Overlap length in
milliseconds. When the sound sequences are mixed back together to form
again a continuous sound stream, this parameter defines how much the
ends of the consecutive sequences will overlap with each other.<br>
<br>
This shouldn't be that critical parameter. If you reduce the
DEFAULT_SEQUENCE_MS setting by a large amount, you might wish to try a
smaller value on this.
</li>
</ul>
<p>Notice that these parameters can also be set during execution time
with functions "<strong>TDStretch::setParameters()</strong>" and "<strong>SoundTouch::setSetting()</strong>".</p>
<p>The table below summaries how the parameters can be adjusted for
different applications:</p>
<table border="1">
<tbody>
<tr>
<td valign="top"><strong>Parameter name</strong></td>
<td valign="top"><strong>Default value magnitude</strong></td>
<td valign="top"><strong>Larger value affects...</strong></td>
<td valign="top"><strong>Smaller value affects...</strong></td>
<td valign="top"><strong>Effect to CPU burden</strong></td>
</tr>
<tr>
<td valign="top">
<pre>SEQUENCE_MS</pre>
</td>
<td valign="top">Default value is relatively large, chosen for
slowing down music tempo</td>
<td valign="top">Larger value is usually better for slowing down
tempo. Growing the value decelerates the "echoing" artifact when
slowing down the tempo.</td>
<td valign="top">Smaller value might be better for speeding up
tempo. Reducing the value accelerates the "echoing" artifact when
slowing down the tempo </td>
<td valign="top">Increasing the parameter value reduces
computation burden</td>
</tr>
<tr>
<td valign="top">
<pre>SEEKWINDOW_MS</pre>
</td>
<td valign="top">Default value is relatively large, chosen for
slowing down music tempo</td>
<td valign="top">Larger value eases finding a good mixing
position, but may cause a "drifting" artifact</td>
<td valign="top">Smaller reduce possibility to find a good mixing
position, but reduce the "drifting" artifact.</td>
<td valign="top">Increasing the parameter value increases
computation burden</td>
</tr>
<tr>
<td valign="top">
<pre>OVERLAP_MS</pre>
</td>
<td valign="top">Default value is relatively large, chosen to
suit with above parameters.</td>
<td valign="top"></td>
<td valign="top">If you reduce the "sequence ms" setting, you
might wish to try a smaller value.</td>
<td valign="top">Increasing the parameter value increases
computation burden</td>
</tr>
</tbody>
</table>
<h3>3.5 Performance Optimizations </h3>
<p><strong>Integer vs floating point:</strong></p>
<p>Floating point sample type is generally recommended because it provides
better sound quality.</p>
<p>However, execution speed difference between integer and floating point processing
depends on the CPU architecture. As rule of thumb,
<ul>
<li>in 32-bit x86 floating point and integer are roughly equally fast</li>
<li>in 64-bit x86/x64 floating point can be significantly faster than integer
version, because MMX integer optimizations are not available in the x64 architecture.
That depends on the compiler however, so that gcc can autovectorize integer routines
to work equally fast as floating point, where as Visual C++ (2017) does not
perform equally well and produces integer code that runs some 3x slower than
SSE-optimized floating poing code.
</li>
<li>in ARMv7 integer routines are twice as fast as floating point. Their
relative difference is roughly the same both with and without NEON; NEON
vfpu can however bring 2.4x speed improvement.
</li>
<li>in other platforms: try out if the execution time performance makes a
big difference</li>
</ul>
</p>
<p><strong>General optimizations:</strong></p>
<p>The time-stretch routine has a 'quick' mode that substantially
speeds up the algorithm but may slightly compromise the sound quality.
This mode is activated by calling SoundTouch::setSetting()
function with parameter id of SETTING_USE_QUICKSEEK and value
"1", i.e. </p>
<blockquote>
<p>setSetting(SETTING_USE_QUICKSEEK, 1);</p>
</blockquote>
<p><strong>CPU-specific optimizations:</strong></p>
<p>Intel x86 specific SIMD optimizations are implemented using compiler
intrinsics, providing about a 3x processing speedup for x86 compatible
processors vs. non-SIMD implementation:</p>
<ul>
<li> MMX optimized routines are used in 32-bit x86 build when 16bit integer
sample type is used</li>
<li> SSE optimized routines are used in 32- and 64-bit x86 CPUs when 32bit
floating point sample type is used</li>
</ul>
<p>The algorithms are tuned to utilize autovectorization efficiently
also in other CPU architectures, for example ARM cpus see approx 2.4x processing
speedup when NEON SIMD support is present.
</p>
<h3>3.5 OpenMP parallel computation</h3>
<p>SoundTouch 1.9 onwards support running the algorithms parallel in several CPU
cores. Based on benchmark the experienced multi-core processing speed-up gain
ranges between +30% (on a high-spec dual-core x86 Windows PC) to 215% (on a moderately low-spec
quad-core ARM of Raspberry Pi2). </p>
<p>See an external blog article with more detailed discussion about the
<a href="http://www.softwarecoven.com/parallel-computing-in-embedded-mobile-devices/">
SoundTouch OpenMP optimization</a>.
</p>
<p>The parallel computing support is implemented using OpenMP spec 3.0
instructions. These instructions are supported by Visual C++ 2008 and later, and
GCC v4.2 and later. Compilers that do not supporting OpenMP will ignore these
optimizations and routines will still work properly. Possible warnings about
unknown #pragmas are related to OpenMP support and can be safely ignored.</p>
<p>The OpenMP improvements are disabled by default, and need to be enabled by
developer during compile-time. Reason for this is that parallel processing adds
moderate runtime overhead in managing the multi-threading, so it may not be
necessary nor desirable in all applications. For example real-time processing
that is not constrained by CPU power will not benefit of speed-up provided by
the parallel processing, in the contrary it may increase power consumption due
to the increased overhead.</p>
<p>However, applications that run on low-spec multi-core CPUs and may otherwise
have possibly constrained performance will benefit of the OpenMP improvements.
This include for example multi-core embedded devices.</p>
<p>OpenMP parallel computation can be enabled before compiling SoundTouch
library as follows:</p>
<ul>
<li><strong>Visual Studio</strong>: Open properties for the <strong>SoundTouch
</strong>sub-project, browse to <strong>C/C++</strong> and <strong>Language
</strong>settings. Set
there "<strong>OpenMP support</strong>" to "<strong>Yes</strong>". Alternatively add
<strong>/openmp</strong> switch to command-line
parameters
</li>
<li><strong>GNU</strong>: Run the configure script with "<strong>./configure
--enable-openmp</strong>" switch, then run make as usually</li>
<li><strong>Android</strong>: Add "<strong>-fopenmp</strong>" switches to compiler & linker
options, see README-SoundTouch-Android.html in the source code package for
more detailed instructions.</li>
</ul>
<hr>
<h2><a name="SoundStretch"></a>4. SoundStretch audio processing utility
</h2>
<p>SoundStretch audio processing utility<br>
Copyright (c) Olli Parviainen 2002-2015</p>
<p>SoundStretch is a simple command-line application that can change
tempo, pitch and playback rates of WAV sound files. This program is
intended primarily to demonstrate how the "SoundTouch" library can be
used to process sound in your own program, but it can as well be used
for processing sound files.</p>
<h3>4.1. SoundStretch Usage Instructions</h3>
<p>SoundStretch Usage syntax:</p>
<blockquote>
<pre>soundstretch infilename outfilename [switches]</pre>
</blockquote>
<p>Where: </p>
<table width="100%" border="0" cellpadding="2">
<tbody>
<tr>
<td valign="top">
<pre>"infilename"</pre>
</td>
<td valign="top">Name of the input sound data file (in .WAV audio
file format). Give "stdin" as filename to use standard input pipe. </td>
</tr>
<tr>
<td valign="top">
<pre>"outfilename"</pre>
</td>
<td valign="top">Name of the output sound file where the
resulting sound is saved (in .WAV audio file format). This parameter
may be omitted if you don't want to save the output (e.g. when
only calculating BPM rate with '-bpm' switch). Give "stdout" as
filename to use standard output pipe.</td>
</tr>
<tr>
<td valign="top">
<pre>[switches]</pre>
</td>
<td valign="top">Are one or more control switches.</td>
</tr>
</tbody>
</table>
<p>Available control switches are:</p>
<table width="100%" border="0" cellpadding="2">
<tbody>
<tr>
<td valign="top">
<pre>-tempo=n </pre>
</td>
<td valign="top">Change the sound tempo by n percents (n = -95.0
.. +5000.0 %) </td>
</tr>
<tr>
<td valign="top">
<pre>-pitch=n</pre>
</td>
<td valign="top">Change the sound pitch by n semitones (n = -60.0
.. + 60.0 semitones) </td>
</tr>
<tr>
<td valign="top">
<pre>-rate=n</pre>
</td>
<td valign="top">Change the sound playback rate by n percents (n
= -95.0 .. +5000.0 %) </td>
</tr>
<tr>
<td valign="top">
<pre>-bpm=n</pre>
</td>
<td valign="top">Detect the Beats-Per-Minute (BPM) rate of the
sound and adjust the tempo to meet 'n' BPMs. When this switch is
applied, the "-tempo" switch is ignored. If "=n" is omitted, i.e.
switch "-bpm" is used alone, then the BPM rate is estimated and
displayed, but tempo not adjusted according to the BPM value. </td>
</tr>
<tr>
<td valign="top">
<pre>-quick</pre>
</td>
<td valign="top">Use quicker tempo change algorithm. Gains speed
but loses sound quality. </td>
</tr>
<tr>
<td valign="top">
<pre>-naa</pre>
</td>
<td valign="top">Don't use anti-alias filtering in sample rate
transposing. Gains speed but loses sound quality. </td>
</tr>
<tr>
<td valign="top">
<pre>-license</pre>
</td>
<td valign="top">Displays the program license text (LGPL)</td>
</tr>
</tbody>
</table>
<p>Notes:</p>
<ul>
<li> To use standard input/output pipes for processing, give "stdin"
and "stdout" as input/output filenames correspondingly. The standard
input/output pipes will still carry the audio data in .wav audio file
format.</li>
<li> The numerical switches allow both integer (e.g. "-tempo=123")
and decimal (e.g. "-tempo=123.45") numbers.</li>
<li> The "-naa" and/or "-quick" switches can be used to reduce CPU
usage while compromising some sound quality</li>
<li> The BPM detection algorithm works by detecting repeating bass or
drum patterns at low frequencies of <250Hz. A lower-than-expected
BPM figure may be reported for music with uneven or complex bass
patterns.</li>
</ul>
<h3>4.2. SoundStretch usage examples </h3>
<p><strong>Example 1</strong></p>
<p>The following command increases tempo of the sound file
"originalfile.wav" by 12.5% and stores result to file
"destinationfile.wav":</p>
<blockquote>
<pre>soundstretch originalfile.wav destinationfile.wav -tempo=12.5</pre>
</blockquote>
<p><strong>Example 2</strong></p>
<p>The following command decreases the sound pitch (key) of the sound
file "orig.wav" by two semitones and stores the result to file
"dest.wav":</p>
<blockquote>
<pre>soundstretch orig.wav dest.wav -pitch=-2</pre>
</blockquote>
<p><strong>Example 3</strong></p>
<p>The following command processes the file "orig.wav" by decreasing
the sound tempo by 25.3% and increasing the sound pitch (key) by 1.5
semitones. Resulting .wav audio data is directed to standard output
pipe:</p>
<blockquote>
<pre>soundstretch orig.wav stdout -tempo=-25.3 -pitch=1.5</pre>
</blockquote>
<p><strong>Example 4</strong></p>
<p>The following command detects the BPM rate of the file "orig.wav"
and adjusts the tempo to match 100 beats per minute. Result is stored
to file "dest.wav":</p>
<blockquote>
<pre>soundstretch orig.wav dest.wav -bpm=100</pre>
</blockquote>
<p><strong>Example 5</strong></p>
<p>The following command reads .wav sound data from standard input pipe
and estimates the BPM rate:</p>
<blockquote>
<pre>soundstretch stdin -bpm</pre>
</blockquote>
<p><strong>Example 6</strong></p>
<p>The following command tunes song from original 440Hz tuning to 432Hz tuning:
this corresponds to lowering the pitch by -0.318 semitones:</p>
<blockquote>
<pre>soundstretch original.wav output.wav -pitch=-0.318</pre>
</blockquote>
<hr>
<h2>5. Change History</h2>
<h3>5.1. SoundTouch library Change History </h3>
<p><b>2.3.1:</b></p>
<ul>
<li>Adjusted cmake build settings and header files that cmake installs</li>
</ul>
<p><b>2.3.0:</b></p>
<ul>
<li>Disable setting "SOUNDTOUCH_ALLOW_NONEXACT_SIMD_OPTIMIZATION" by default. The original
purpose of this setting was to avoid performance penalty due to unaligned SIMD memory
accesses in old CPUs, but that is not any more issue in concurrent CPU SIMD implementations
and having this setting enabled can cause slight compromise in result quality.
</li>
<li>Bugfix: soundtouch.clear() to really clear whole processing pipeline state. Earlier
individual variables were left uncleared, which caused slightly different result if
the same audio stream were processed again after calling clear().
</li>
<li>Bugfix: TDstretch to align initial offset position to be in middle of correlation search
window. This ensures that with zero tempo change the output will be same as input.
</li>
<li>Bugfix: Fix a bug in TDstrectch with too small initial skipFract value that occurred
with certain processing parameter settings: Replace assert with assignment that
corrects the situation.
</li>
<li>Remove OpenMP "_init_threading" workaround from Android build as it's not needed with concurrent
Android SDKs any more.</li>
</ul>
<p><b>2.2:</b></p>
<ul>
<li>Improved source codes so that compiler can autovectorize them more effectively.
This brings remarkable improvement e.g. ARM cpus equipped with NEON vfpu: Bencmarked
2.4x improvement in execution speed in ARMv7l vs the previous SoundTouch version
for both integer and floating point sample types.
</li>
<li>Bugfix: Resolved bad sound quality when using integer sample types in non-x86 CPU</li>
<li>Bugfix: Fixed possible reading past end of array in BPM peak detection algorithm</li>
</ul>
<p><b>2.1.2:</b></p>
<ul>
<li>Bump version to 2.1.2 also in configure.ac. The earlier release had old version info for GNU autotools.</li>
</ul>
<p><b>2.1.1:</b></p>
<ul>
<li>Bugfixes: Fixed potential buffer overwrite bugs in WavFile routines. Replaced asserts with runtime exceptions.
</li>
<li>Android: Migrated the SoundTouch Android example to new Android Studio</li>
<li>Automake: unset ACLOCAL in bootstrap script in case earlier build script has set it</li>
</ul>
<p><b>2.1:</b></p>
<ul>
<li>Refactored C# interface example</li>
<li>Disable anti-alias filter when switch
SOUNDTOUCH_PREVENT_CLICK_AT_RATE_CROSSOVER defined because anti-alias
filter cause slight click if the rate change crosses zero during
processing</li>
<li>Added script for building SoundTouchDll dynamic-link-library for GNU platforms</li>
<li>Rewrote Beats-per-Minute analysis algorithm for more reliable BPM detection</li>
<li>Added BPM functions to SoundTouchDll API</li>
<li>Migrated Visual Studio project files to MSVC 201x format</li>
<li>Replaced function parameter value asserts with runtime exceptions</li>
<li>Code maintenance & style cleanup</li>
</ul>
<p><b>2.0:</b></p>
<ul>
<li>Added functions to get initial processing latency, duration ratio between the original input and processed
output tracks, and clarified reporting of input/output batch sizes</li>
<li>Fixed issue that added brief sequence of silence to beginning of output audio</li>
<li>Adjusted algorithm parameters to reduce reverberating effect at tempo slowdown</li>
<li>Bugfix: Fixed a glitch that could cause negative array indexing in quick seek algorithm</li>
<li>Bugfix: flush() didn't properly flush final samples from the pipeline on 2nd time in case that soundtouch
object instance was recycled and used for processing a second audio stream.</li>
<li>Bugfix: Pi value had incorrect 9th/10th decimals</li>
<li>Added C# example application that uses SoundTouch dll library for processing MP3 files</li>
</ul>
<p><b>1.9.2:</b></p>
<ul>
<li>Fix in GNU package configuration</li>
</ul>
<p><b>1.9.1:</b></p>
<ul>
<li>Improved SoundTouch::flush() function so that it returns precisely the desired amount of samples for exact
output duration control</li>
<li>Redesigned quickseek algorithm for improved sound quality when using the quickseek mode. The new quickseek
algorithm can find 99% as good results as the
default full-scan mode, while the quickseek algorithm is remarkable less
CPU intensive.</li>
<li>Added adaptive integer divider scaling for improved sound quality when using integer processing algorithm
</li>
</ul>
<p><b>1.9:</b></p>
<ul>
<li>Added support for parallel computation support via OpenMP primitives for better performance in multicore
systems.
Benchmarks show that achieved parallel processing speedup improvement
typically range from +30% (x86 dual-core) to +180% (ARM quad-core). The
OpenMP optimizations are disabled by default, see OpenMP notes above in this
readme file how to enabled these optimizations.</li>
<li>Android: Added support for Android devices featuring X86 and MIPS CPUs,
in addition to ARM CPUs.</li>
<li>Android: More versatile Android example application that processes WAV
audio files with SoundTouch library</li>
<li>Replaced Windows-like 'BOOL' types with native 'bool'</li>
<li>Changed documentation token to "dist_doc_DATA" in Makefile.am file</li>
<li>Miscellaneous small fixes and improvements</li>
</ul>
<p><b>1.8.0:</b></p>
<ul>
<li>Added support for multi-channel audio processing</li>
<li>Added support for <b>cubic</b> and <b>shannon</b> interpolation for rate and pitch shift effects besides
the original <b>linear</b> interpolation, to reduce aliasing at high frequencies due to interpolation.
Cubic interpolation is used as default for floating point processing, and linear interpolation for integer
processing.</li>
<li>Fixed bug in anti-alias filtering that limited stop-band attenuation to -10 dB instead of <-50dB, and
increased filter length from 32 to 64 taps to further reduce aliasing due to frequency folding.</li>
<li>Performance improvements in cross-correlation algorithm</li>
<li>Other bug and compatibility fixes</li>
</ul>
<p><b>1.7.1:</b></p>
<ul>
<li>Added files for Android compilation
</ul>
<p><b>1.7.0:</b></p>
<ul>
<li>Sound quality improvements/li>
<li>Improved flush() to adjust output sound stream duration to match better with
ideal duration</li>
<li>Rewrote x86 cpu feature check to resolve compatibility problems</li>
<li>Configure script automatically checks if CPU supports mmx & sse compatibility for GNU platform, and
the script support now "--enable-x86-optimizations" switch to allow disabling x86-specific optimizations.</li>
<li>Revised #define conditions for 32bit/64bit compatibility</li>
<li>gnu autoconf/automake script compatibility fixes</li>
<li>Tuned beat-per-minute detection algorithm</li>
</ul>
<p><b>1.6.0:</b></p>
<ul>
<li> Added automatic cutoff threshold adaptation to beat detection
routine to better adapt BPM calculation to different types of music</li>
<li> Retired 3DNow! optimization support as 3DNow! is nowadays
obsoleted and assembler code is nuisance to maintain</li>
<li>Retired "configure" file from source code package due to
autoconf/automake versio conflicts, so that it is from now on to be
generated by invoking "boostrap" script that uses locally available
toolchain version for generating the "configure" file</li>
<li>Resolved namespace/label naming conflicts with other libraries by
replacing global labels such as INTEGER_SAMPLES with more specific
SOUNDTOUCH_INTEGER_SAMPLES etc.<br>
</li>
<li>Updated windows build scripts & project files for Visual
Studio 2008 support</li>
<li> Updated SoundTouch.dll API for .NET compatibility</li>
<li> Added API for querying nominal processing input & output
sample batch sizes</li>
</ul>
<p><strong>1.5.0:</strong></p>
<ul>
<li> Added normalization to correlation calculation and improvement
automatic seek/sequence parameter calculation to improve sound quality</li>
<li> Bugfixes:
<ul>
<li> Fixed negative array indexing in quick seek algorithm</li>
<li> FIR autoalias filter running too far in processing buffer</li>
<li> Check against zero sample count in rate transposing</li>
<li> Fix for x86-64 support: Removed pop/push instructions from
the cpu detection algorithm.</li>
<li> Check against empty buffers in FIFOSampleBuffer</li>
<li> Other minor fixes & code cleanup</li>
</ul>
</li>
<li> Fixes in compilation scripts for non-Intel platforms</li>
<li> Added Dynamic-Link-Library (DLL) version of SoundTouch library
build, provided with Delphi/Pascal wrapper for calling the dll routines
</li>
<li> Added #define PREVENT_CLICK_AT_RATE_CROSSOVER that prevents a
click artifact when crossing the nominal pitch from either positive to
negative side or vice versa</li>
</ul>
<p><strong>1.4.1:</strong></p>
<ul>
<li> Fixed a buffer overflow bug in BPM detect algorithm routines if
processing more than 2048 samples at one call</li>
</ul>
<p><strong>1.4.0:</strong></p>
<ul>
<li> Improved sound quality by automatic calculation of time stretch
algorithm processing parameters according to tempo setting</li>
<li> Moved BPM detection routines from SoundStretch application into
SoundTouch library</li>
<li> Bugfixes: Usage of uninitialied variables, GNU build scripts,
compiler errors due to 'const' keyword mismatch.</li>
<li> Source code cleanup</li>
</ul>
<p><strong>1.3.1: </strong> </p>
<ul>
<li> Changed static class declaration to GCC 4.x compiler compatible
syntax.</li>
<li> Enabled MMX/SSE-optimized routines also for GCC compilers.
Earlier the MMX/SSE-optimized routines were written in
compiler-specific inline assembler, now these routines are migrated to
use compiler intrinsic syntax which allows compiling the same
MMX/SSE-optimized source code with both Visual C++ and GCC compilers.</li>
<li> Set floating point as the default sample format and added switch
to the GNU configure script for selecting the other sample format.</li>
</ul>
<p><strong>1.3.0: </strong> </p>
<ul>
<li> Fixed tempo routine output duration inaccuracy due to rounding
error</li>
<li> Implemented separate processing routines for integer and
floating arithmetic to allow improvements to floating point routines
(earlier used algorithms mostly optimized for integer arithmetic also
for floating point samples)</li>
<li> Fixed a bug that distorts sound if sample rate changes during
the sound stream</li>
<li> Fixed a memory leak that appeared in MMX/SSE/3DNow! optimized
routines</li>
<li> Reduced redundant code pieces in MMX/SSE/3DNow! optimized
routines vs. the standard C routines.</li>
<li> MMX routine incompatibility with new gcc compiler versions</li>
<li> Other miscellaneous bug fixes</li>
</ul>
<p><strong>1.2.1: </strong> </p>
<ul>
<li> Added automake/autoconf scripts for GNU platforms (in courtesy
of David Durham)</li>
<li> Fixed SCALE overflow bug in rate transposer routine.</li>
<li> Fixed 64bit address space bugs.</li>
<li> Created a 'soundtouch' namespace for SAMPLETYPE definitions.</li>
</ul>
<p><strong>1.2.0: </strong> </p>
<ul>
<li> Added support for 32bit floating point sample data type with
SSE/3DNow! optimizations for Win32 platform (SSE/3DNow! optimizations
currently not supported in GCC environment)</li>
<li> Replaced 'make-gcc' script for GNU environment by master
Makefile</li>
<li> Added time-stretch routine configurability to SoundTouch main
class</li>
<li> Bugfixes</li>
</ul>
<p><strong>1.1.1: </strong> </p>
<ul>
<li> Moved SoundTouch under lesser GPL license (LGPL). This allows
using SoundTouch library in programs that aren't released under GPL
license.</li>
<li> Changed MMX routine organiation so that MMX optimized routines
are now implemented in classes that are derived from the basic classes
having the standard non-mmx routines.</li>
<li> MMX routines to support gcc version 3.</li>
<li> Replaced windows makefiles by script using the .dsw files</li>
</ul>
<p><strong>1.0.1: </strong> </p>
<ul>
<li> "mmx_gcc.cpp": Added "using namespace std" and removed "return
0" from a function with void return value to fix compiler errors when
compiling the library in Solaris environment.</li>
<li> Moved file "FIFOSampleBuffer.h" to "include" directory to allow
accessing the FIFOSampleBuffer class from external files.</li>
</ul>
<p><strong>1.0: </strong> </p>
<ul>
<li> Initial release</li>
</ul>
<h3>5.2. SoundStretch application Change History </h3>
<p><b>1.9:</b></p>
<ul>
<li>Added support for WAV file 'fact' information chunk.</li>
</ul>
<p><b>1.7.0:</b></p>
<ul>
<li>Bugfixes in Wavfile: exception string formatting, avoid getLengthMs() integer
precision overflow, support WAV files using 24/32bit sample format.</li>
</ul>
<p><b>1.5.0:</b></p>
<ul>
<li> Added "-speech" switch to activate algorithm parameters more
suitable for speech processing than the default parameters tuned for
music processing.</li>
</ul>
<p><strong>1.4.0:</strong></p>
<ul>
<li> Moved BPM detection routines from SoundStretch application into
SoundTouch library</li>
<li> Allow using standard input/output pipes as audio processing
input/output streams</li>
</ul>
<p><strong>1.3.0:</strong></p>
<ul>
<li> Simplified accessing WAV files with floating point sample
format.</li>
</ul>
<p><strong>1.2.1: </strong> </p>
<ul>
<li> Fixed 64bit address space bugs.</li>
</ul>
<p><strong>1.2.0: </strong> </p>
<ul>
<li> Added support for 32bit floating point sample data type</li>
<li> Restructured the BPM routines into separate library</li>
<li> Fixed big-endian conversion bugs in WAV file routines (hopefully
:)</li>
</ul>
<p><strong>1.1.1: </strong> </p>
<ul>
<li> Fixed bugs in WAV file reading & added byte-order conversion
for big-endian processors.</li>
<li> Moved SoundStretch source code under 'example' directory to
highlight difference from SoundTouch stuff.</li>
<li> Replaced windows makefiles by script using the .dsw files</li>
<li> Output file name isn't required if output isn't desired (e.g. if
using the switch '-bpm' in plain format only)</li>
</ul>
<p><strong>1.1:</strong></p>
<ul>
<li> Fixed "Release" settings in Microsoft Visual C++ project file
(.dsp)</li>
<li> Added beats-per-minute (BPM) detection routine and command-line
switch "-bpm"</li>
</ul>
<p><strong>1.01: </strong> </p>
<ul>
<li> Initial release</li>
</ul>
<hr>
<h2>6. Acknowledgements </h2>
<p>Kudos for these people who have contributed to development or
submitted bugfixes:</p>
<ul>
<li> Arthur A</li>
<li> Paul Adenot</li>
<li> Richard Ash</li>
<li> Stanislav Brabec</li>
<li> Christian Budde</li>
<li> Jamie Bullock</li>
<li> Chris Bryan</li>
<li> Jacek Caban</li>
<li> Marketa Calabkova</li>
<li> Brian Cameron</li>
<li> Jason Champion</li>
<li> Giuseppe Cigala</li>
<li> David Clark</li>
<li> Patrick Colis</li>
<li> Miquel Colon</li>
<li> Jim Credland</li>
<li> Sandro Cumerlato</li>
<li> Gerry Fan</li>
<li> Justin Frankel</li>
<li> Masa H.</li>
<li> Jason Garland</li>
<li> Takashi Iwai</li>
<li> Thomas Klausner</li>
<li> Lu Zhihe</li>
<li> Luzpaz</li>
<li> Tony Mechelynck </li>
<li> Mathias Möhl</li>
<li> Yuval Naveh</li>
<li> Mats Palmgren </li>
<li> Chandni Patel</li>
<li> Paulo Pizarro</li>
<li> Andrey Ponomarenko</li>
<li> Blaise Potard</li>
<li> Michael Pruett</li>
<li> Rajeev Puran</li>
<li> RJ Ryan</li>
<li> John Sheehy</li>
<li> Tim Shuttleworth</li>
<li> Albert Sirvent</li>
<li> Tyson Smith</li>
<li> John Stumpo</li>
<li> Mario di Vece</li>
<li> Rémi Verschelde</li>
<li> Katja Vetter</li>
<li> Wu Q.</li>
</ul>
<p>Moral greetings to all other contributors and users also!</p>
<hr>
<h2>7. LICENSE </h2>
<p>SoundTouch audio processing library<br>
Copyright (c) Olli Parviainen</p>
<p>This library is free software; you can redistribute it and/or modify
it under the terms of the GNU Lesser General Public License version 2.1
as published by the Free Software Foundation.</p>
<p>This library is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser
General Public License for more details.</p>
<p>You should have received a copy of the GNU Lesser General Public
License along with this library; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA</p>
<p>---</p>
<p>commercial license alternative also available, contact author for details.</p>
<hr>
</body>
</html>