2018-09-08 18:15:39 +02:00
|
|
|
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
|
|
|
|
|
<html>
|
2020-10-15 17:20:52 +02:00
|
|
|
|
2018-09-08 18:15:39 +02:00
|
|
|
<head>
|
|
|
|
|
<title>SoundTouch library README</title>
|
|
|
|
|
<meta http-equiv="Content-Language" content="en-us">
|
|
|
|
|
<meta name="author" content="Olli Parviainen">
|
2020-10-15 17:20:52 +02:00
|
|
|
<meta name="description" content="Readme file for SoundTouch audio processing library">
|
|
|
|
|
<style>
|
|
|
|
|
body {
|
|
|
|
|
font-family: Arial, Helvetica;
|
|
|
|
|
}
|
|
|
|
|
</style>
|
2018-09-08 18:15:39 +02:00
|
|
|
</head>
|
2020-10-15 17:20:52 +02:00
|
|
|
|
2018-09-08 18:15:39 +02:00
|
|
|
<body class="normal">
|
2020-10-15 17:20:52 +02:00
|
|
|
<hr>
|
2024-03-29 19:49:53 +01:00
|
|
|
<h1>SoundTouch audio processing library v2.3.3</h1>
|
2024-02-11 14:27:45 +01:00
|
|
|
<p class="normal">SoundTouch library Copyright © Olli Parviainen 2001-2024</p>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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
|
2021-10-14 17:29:15 +02:00
|
|
|
href="https://codeberg.org/soundtouch/soundtouch.git">https://codeberg.org/soundtouch/soundtouch.git</a></p>
|
2020-10-15 17:20:52 +02:00
|
|
|
<hr>
|
|
|
|
|
<h2>2. Compiling SoundTouch</h2>
|
|
|
|
|
<p>Before compiling, notice that you can choose the sample data format if it's
|
2021-08-21 10:38:36 +02:00
|
|
|
desirable to use 16bit integer sample data instead of floating point samples. See
|
2020-10-15 17:20:52 +02:00
|
|
|
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
|
2018-09-08 18:15:39 +02:00
|
|
|
dynamic-load library for processing mp3 audio.
|
2020-10-15 17:20:52 +02:00
|
|
|
<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>
|
2021-01-30 18:02:08 +01:00
|
|
|
<p>In other VC++ versions the required library will be expectedly found in similar
|
2020-10-15 17:20:52 +02:00
|
|
|
"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,
|
2021-08-21 10:38:36 +02:00
|
|
|
thus choose the proper version to allow the program to start.</p>
|
2020-10-15 17:20:52 +02:00
|
|
|
<h3>2.2. Building in Gnu platforms</h3>
|
|
|
|
|
<p>The SoundTouch library compiles in practically any platform
|
2021-08-21 10:38:36 +02:00
|
|
|
supporting GNU compiler (GCC) tools.
|
|
|
|
|
<h4>2.2.1 Compiling with autotools</h4>
|
|
|
|
|
<p>To install build prerequisites for 'autotools' tool chain:</p>
|
2021-01-30 18:02:08 +01:00
|
|
|
<pre> sudo apt-get install automake autoconf libtool build-essential</pre>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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>
|
2018-09-08 18:15:39 +02:00
|
|
|
|
2021-08-21 10:38:36 +02:00
|
|
|
<b>Compiling portable Shared Library / DLL version</b>
|
2022-10-31 17:51:35 +01:00
|
|
|
<p> The GNU autotools compilation automatically builds an additional dynamic-link version
|
|
|
|
|
of SoundTouch library that features position-independent code and "C"-style API that is
|
|
|
|
|
more suitable for calling the SoundTouch routines from other programming languages.</p>
|
|
|
|
|
<p>This dynamic-link library is built under source/SoundTouchDLL directory, whose
|
|
|
|
|
subdirectories also comtain simple example apps that use the dynamic-link library.
|
|
|
|
|
</p>
|
2018-09-08 18:15:39 +02:00
|
|
|
|
2021-08-21 10:38:36 +02:00
|
|
|
<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>
|
2024-03-29 18:32:39 +01:00
|
|
|
<p>To list available build options:</p>
|
|
|
|
|
<pre>
|
|
|
|
|
cmake -LH</pre>
|
2021-08-21 10:38:36 +02:00
|
|
|
<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>
|
|
|
|
|
|
2020-10-15 17:20:52 +02:00
|
|
|
<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>"
|
2018-09-08 18:15:39 +02:00
|
|
|
in the source code package. </p>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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>
|
2018-09-08 18:15:39 +02:00
|
|
|
|
2020-10-15 17:20:52 +02:00
|
|
|
<h3>2.4. Building in Mac</h3>
|
|
|
|
|
<p>Install autoconf tool as instructed in <a
|
2021-08-21 10:38:36 +02:00
|
|
|
href="http://macappstore.org/autoconf/">http://macappstore.org/autoconf/</a>, or alternatively the 'cmake' toolchain.</p>
|
2020-10-15 17:20:52 +02:00
|
|
|
<p>Then, build as described above in section "Building in Gnu platforms".</p>
|
2019-10-28 18:04:28 +01:00
|
|
|
|
2020-10-15 17:20:52 +02:00
|
|
|
<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
|
2021-01-30 18:02:08 +01:00
|
|
|
smaller value on this.
|
|
|
|
|
</li>
|
2020-10-15 17:20:52 +02:00
|
|
|
</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/">
|
2021-01-30 18:02:08 +01:00
|
|
|
SoundTouch OpenMP optimization</a>.
|
|
|
|
|
</p>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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
|
2021-01-30 18:02:08 +01:00
|
|
|
parameters
|
|
|
|
|
</li>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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>
|
2024-02-11 14:27:45 +01:00
|
|
|
Copyright (c) Olli Parviainen 2002-2024</p>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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>
|
2019-01-07 17:55:36 +01:00
|
|
|
<ul>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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>
|
2024-03-29 19:49:53 +01:00
|
|
|
<p><b>2.3.3:</b></p>
|
2024-02-12 16:48:11 +01:00
|
|
|
<ul class="current">
|
2024-03-29 19:49:53 +01:00
|
|
|
<li>Fixing compiler warnings, maintenance fixes to make/build files for various systems
|
|
|
|
|
</li>
|
|
|
|
|
</ul>
|
|
|
|
|
<p><b>2.3.2:</b></p>
|
|
|
|
|
<ul>
|
|
|
|
|
<li>Improve autotools makefiles to build the `SoundTouchDLL` dynamic-link link library with
|
2022-11-08 17:02:17 +01:00
|
|
|
C-style API. This library variation is easier to import and use from other programming
|
|
|
|
|
languages than the default C++ library.
|
|
|
|
|
</li>
|
|
|
|
|
</ul>
|
2021-09-01 17:46:54 +02:00
|
|
|
<p><b>2.3.1:</b></p>
|
|
|
|
|
<ul>
|
|
|
|
|
<li>Adjusted cmake build settings and header files that cmake installs</li>
|
|
|
|
|
</ul>
|
2021-08-21 10:38:36 +02:00
|
|
|
<p><b>2.3.0:</b></p>
|
2021-01-30 18:02:08 +01:00
|
|
|
<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>
|
2021-08-21 10:38:36 +02:00
|
|
|
<li>Bugfix: Fix a bug in TDstrectch with too small initial skipFract value that occurred
|
2022-10-31 17:51:35 +01:00
|
|
|
with certain processing parameter settings: Replace assert with assignment that
|
2021-08-21 10:38:36 +02:00
|
|
|
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>
|
2021-01-30 18:02:08 +01:00
|
|
|
</ul>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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>
|
2019-01-07 17:55:36 +01:00
|
|
|
<li>Bugfix: Fixed possible reading past end of array in BPM peak detection algorithm</li>
|
|
|
|
|
</ul>
|
2018-12-02 09:43:00 +01:00
|
|
|
<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>
|
2018-11-14 18:25:34 +01:00
|
|
|
<p><b>2.1.1:</b></p>
|
|
|
|
|
<ul>
|
2020-10-15 17:20:52 +02:00
|
|
|
<li>Bugfixes: Fixed potential buffer overwrite bugs in WavFile routines. Replaced asserts with runtime exceptions.
|
|
|
|
|
</li>
|
2018-11-14 18:25:34 +01:00
|
|
|
<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>
|
2018-09-08 18:15:39 +02:00
|
|
|
<p><b>2.1:</b></p>
|
|
|
|
|
<ul>
|
|
|
|
|
<li>Refactored C# interface example</li>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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>
|
2018-09-08 18:15:39 +02:00
|
|
|
<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>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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>
|
2018-09-08 18:15:39 +02:00
|
|
|
</ul>
|
|
|
|
|
<p><b>1.9.2:</b></p>
|
|
|
|
|
<ul>
|
2020-10-15 17:20:52 +02:00
|
|
|
<li>Fix in GNU package configuration</li>
|
2018-09-08 18:15:39 +02:00
|
|
|
</ul>
|
|
|
|
|
<p><b>1.9.1:</b></p>
|
|
|
|
|
<ul>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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
|
2018-09-08 18:15:39 +02:00
|
|
|
CPU intensive.</li>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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
|
2018-09-08 18:15:39 +02:00
|
|
|
the original <b>linear</b> interpolation, to reduce aliasing at high frequencies due to interpolation.
|
2020-10-15 17:20:52 +02:00
|
|
|
Cubic interpolation is used as default for floating point processing, and linear interpolation for integer
|
2018-09-08 18:15:39 +02:00
|
|
|
processing.</li>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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
|
2018-09-08 18:15:39 +02:00
|
|
|
ideal duration</li>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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>
|
2024-02-11 14:27:45 +01:00
|
|
|
<p><b>2.3.3:</b></p>
|
2024-02-12 16:48:11 +01:00
|
|
|
<ul class="current_soundstretch">
|
2024-02-11 14:27:45 +01:00
|
|
|
<li>Added support for Asian / non-latin filenames in Windows. Gnu platform has supported them already earlier.</li>
|
|
|
|
|
</ul>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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
|
2018-09-08 18:15:39 +02:00
|
|
|
precision overflow, support WAV files using 24/32bit sample format.</li>
|
2020-10-15 17:20:52 +02:00
|
|
|
</ul>
|
2018-09-08 18:15:39 +02:00
|
|
|
<p><b>1.5.0:</b></p>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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>
|
2023-03-19 16:06:47 +01:00
|
|
|
<li> Serge Sans Paille</li>
|
2020-10-15 17:20:52 +02:00
|
|
|
<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>
|
2018-09-08 18:15:39 +02:00
|
|
|
</body>
|
2020-10-15 17:20:52 +02:00
|
|
|
|
2021-01-30 18:02:08 +01:00
|
|
|
</html>
|
