Alexander Kojevnikov | Blog

Faster Fast Fourier Transform

2014-08-31T00:00:00+00:00

Spek is already pretty fast at analyzing audio files and creating their spectrograms. Apparently, it can be made even faster!

A sizable chunk of processing time is spent calculating discrete Fourier transforms of audio samples. In this post I'm going to explore a few DFT libraries and compare their performance to what Spek is currently using.

This won't be a comprehensive overview of all available libraries. I'm going to benchmark only what Spek needs:

Relatively small, power of 2 transform sizes. The default size in Spek is 2¹¹, this will be adjustable in the future, but not by much.
Single precision floating point numbers. 24-bit significand is enough when working with audio samples.
Only one-dimensional forward real transforms.
Last but not least, the DFT library should be free software, actively maintained, fairly popular and multi-platform.

The list of candidates is surprisingly short:

avfft, which is part of the FFmpeg project. This is what Spek uses.
FFTW. Probably the most respectable FFT library of all. Supports both in-place and out-of-place transforms.
djbfft by Dr. Bernstein of daemontools fame. It's not as performance tuned as the other two (e.g. it doesn't use SIMD instructions), but I want to see how it fares against them.

The test setup is straightforward (the code is on GitHub in case you're curious):

Generate pseudo-random single-precision floating point samples. The number of samples corresponds to 10 minutes of signal at standard 44.1 kHz sampling rate.
For all libraries and for all transform sizes from 2⁹ to 2¹³ run non-overlapping FFTs for the entire signal.
Measure wall clock time only for FFTs (exclude sample generation and library initialization), take 5 measurements for every combination, report the fastest.

Results:

As expected, djbfft is the slowest, using SIMD instructions does make things faster. I am however surprised how fast the FFTW is compared to avfft (approximately 60% faster!)

Conclusion: Next version of Spek will switch to FFTW.

Spek 0.8.0 Released

2012-09-23T00:00:00+00:00

After more than a year in the baking, a new version of Spek is out!

This release is almost a complete rewrite driven by the switch from GTK+ to wxWidgets to simplify packaging and to improve integration on Windows and Mac OS X. The switch also allowed to make a singe-exe version of Spek on Windows, which was a frequently requested feature.

Spek now allows to change the spectral density range, which is essential when trying to detect lossy transcodes. It also handles the low end of the density better resulting in less noisy spectrograms.

There were also some infrastructure changes: downloads and issues have been moved from Google Code to GitHub, and wiki pages with platform-specific installation instructions have been moved to a single INSTALL file.

Read the changelog for the full list of changes in this release. Download links and installation instructions are available on the Spek homepage, get it while it's hot!

Spek and Lion

2011-08-17T00:00:00+00:00

Spek is now fully compatible with the recently released Mac OS X Lion. The bug that prevented it from starting up is now fixed, thanks to John Ralls and the awesome GTK-OSX project.

Get your copy of Spek today!

Spek 0.7 Released

2011-04-24T00:00:00+00:00

I'm happy to announce the release of Spek 0.7 – a multi-platform acoustic spectrum analyser.

This version features multi-lingual support and a much better OS X integration. It also includes performance tweaks and many bug fixes.

Read the NEWS file for a full change log.

Download links and installation instructions are on the Spek website, get it while it's hot!

Spek status update

2011-03-16T00:00:00+00:00

Just a quick update on the status of the project.

Spek is now translatable, the next version will feature at least these translations: Dutch, German, Russian, Swedish and Ukrainian. We use Transifex to manage translations. Feel free to translate it into your language or to improve an existing translation. Transifex is a very easy to use tool, you don't have to know anything about programming to translate.

Some progress has been made on the packaging front, Spek packages are now available in Arch Linux, FreeBSD and Gentoo repositories.

I also created a package for Debian which needs a sponsor. If you are a Debian Developer I would appreciate a review and eventually an upload. The package can be found on mentors.debian.net.

Last but not least, the next version will show some OS X love. Many OS X specific bugs have already been fixed or are scheduled to be fixed for the next release.

I expect to release Spek 0.7 sometime this spring. If you are feeling adventurous, try Spek from git master, compilation instructions are on the wiki.

Spek 0.6 Released

2010-07-13T00:00:00+00:00

I'm happy to announce the release of Spek 0.6 – a multi-platform acoustic spectrum analyser.

This version is about 3 times faster than 0.5 thanks to the lightning-fast FFmpeg decoders and the new multi-threaded analysis algorithm.

Spek 0.6 also features dramatically reduced size of the Windows installer (from 17.1 MiB to 9.8 MiB) and OS X bundle (from 10.5 MiB to 6.1 MiB)

Read the NEWS file for a complete change log.

GStreamer, FFmpeg and Spek

2010-07-04T00:00:00+00:00

Next version of Spek will use FFmpeg libraries to decode audio files. There are several reasons for the switch from GStreamer:

GStreamer is a fantastic framework for building complex multimedia pipelines, however what Spek really needs is a simple decoder and FFmpeg's libavformat and libavcodec do just that.

To handle some audio formats (e.g. APE and DTS), GStreamer relies on FFmpeg anyway, so the switch will result in fewer dependencies. It doesn't matter too much on GNU/Linux, but this will reduce the size of the Windows and Mac OS X installers.

Spek used GStreamer's spectrum plugin to perform the actual spectral analysis, with FFmpeg I had to implement it myself. The code I ended up with is very compact and gives room for a lot of experimentation, from using different window functions (it's still Hamming) and working on performance optimisations to switching to a faster FFT library.

The last bit is actually done, Spek now uses FFTW which in my tests is 1.5x to 2x faster than Kiss FFT used by GStreamer. Apart from that, FFTW can scale to multiple threads with near linear performance increase, future versions of Spek will take advantage of this.

UPDATE: As one of commenters pointed out, FFTs on small number of samples are not very parallelisable and my benchmarks confirm this. Also, I switched from FFTW to avfft which is built into FFmpeg. It's a little bit faster than FFTW for my particular use case. Lastly, 1.5x to 2x speed up was actually caused by a faster decoder, not by a faster FFT library.

Another thing that would be hard with GStreamer is static cross-compilation using mingw-cross-env to produce a single Windows executable. Because FFmpeg doesn't employ a plugin architecture, static linking is not an issue.

Last, but not least, the whole experience was very educational. I now remember why I loved C back in the days, the trick is to not even attempt to write any GObject code with it, that's what Vala is for. Fast Fourier Transform and the maths behind it is much fun, and NR was very helpful here.

For the curious, the code is merged to git master and pushed to Gitorious. The next version of Spek will be released sometime in July.

Spek 0.5 - Mac OS X port

2010-06-27T00:00:00+00:00

I'm happy to announce the release of version 0.5 of Spek, a multi-platform acoustic spectrum analyser.

Changes since the previous release:

Significantly speed up spectral analysis by using the optimal number of frequency bands.
DTS files support.
Distribute Windows version as a ZIP archive in addition to the MSI installer.
Mac OS X installer.
Use Pango to render text.
Brand new icon.

This is the first version of Spek that features a Mac OS X port. All binaries in the app bundle are compiled completely from scratch using a slightly modified version of Aaron's excellent bockbuild project (the same tool that is used to package Banshee for OS X).

If your project uses the GNOME stack and you want to port it to Mac OS X -- give bockbuild a try. Check Spek's bundle script for gory details.

Spek 0.4 Released

2010-05-21T00:00:00+00:00

I've just released version 0.4 of Spek -- an acoustic spectrum analyser for GNU/Linux and Windows.

This version brings the following features and improvements:

Associate with audio files ("Open with..." menu in file managers.)
Show the name of the open file in the window title.
Support 24-bit FLACs.
Drag and Drop support.
Show file name and its properties in the window.

Grab Spek 0.4 while it's hot, the source code tarball and the Windows installer can be downloaded here.

A special "thank you" to users of a popular private tracker, their great suggestions helped to shape this release.

P.S. Hello, Planet GNOME! :)

Spek 0.3 Released

2010-05-17T00:00:00+00:00

Spek 0.3 is out, just 2 days after the previous release. This version includes the following features and fixes:

Horizontal (time) and vertical (frequency) rulers, auto-adjustable to the duration and sample rate of the audio file and to the window size.
First take on the command line arguments parsing, for now Spek understands only --help and --version (in addition to the usual gtk+ and gst options.)
Open the file passed as an argument. In the future Spek will also associate itself with common audio formats so that you could use it as a viewer.
Fix a crash when the window width is very small.

Spek 0.3 source code tarball and the Windows installer can be downloaded here.

Spek 0.2 - now works in Windows

2010-05-15T00:00:00+00:00

Spek now has a fully functional Windows port! The installer bundles GTK+, GStreamer and all their dependencies, you don't need to install anything else.

I also moved project hosting over to Google Code so Spek now has a full-fledged bug tracker. The source code is still hosted on Gitorious.

If you are tired of starting up monstrous applications like Adobe Audition or Audacity just to check a file's spectrogram, give Spek a shot!

Introducing Spek

2010-05-10T00:00:00+00:00

I just released version 0.1 of Spek -- a little program that shows spectrograms of audio files. Spek is written in Vala and uses the standard GNOME stack: GLib, GTK+, Cairo and GStreamer.

Spectrograms are used to analyse the quality of audio files, you can easily detect lossy re-encodes, web-rips and other badness by just looking at the spectrogram.

This version of Spek doesn't do much apart from showing the actual spectrogram and allowing to save it as a PNG image. However, I plan to add a bunch of features before releasing version 1.0:

Windows port.
Horizontal (time) and vertical (frequency) rulers.
Zooming, scrolling and other adjustments.
Associate Spek with audio files to use it as a viewer.
Use multiple threads to speed-up processing.

You can download the tarball from the project's website. To build and run:

% tar -xjvf spek-0.1.tar.bz2
% cd spek-0.1
% ./configure
% make
% src/spek

Or sudo make install to have it installed.

Contribute

The code is available on Gitorious. I really need and will appreciate help in these areas:

Packaging
Translations
Review of autotools-related code

Why Vala?

I initially wanted to write Spek in Haskell to practice the language after reading The Haskell School of Expression (generously sent to me by Jorge). However, after writing a bit of code I realised that all the functional goodness of Haskell is not used at all because Spek is simply a bit of glue between GTK+ and GStreamer, with very little code of its own. Also, the prototype's executable size of 14+ MiB didn't help much in convincing myself that Haskell was a good pick for this project.

I was left with C and C#. The latter didn't feel like a good idea for the same reasons: I wanted something lightweight for this small little app, C# would require a lot of dependencies. Also, while I really like C# as a language and Mono/.NET as a platform, I wanted to try something new for a change -- I already use them full-time on my day job and when hacking Banshee.

I wrote very little C/C++ code since late 90's, now I know why I didn't miss it much -- it's so incredibly verbose! After a few hours I gave up, decided to do some research and found Vala :)

The impression so far is hugely positive. Vala still has a few rough edges but nothing too bad and not work-aroundable. If you are tired of using C for your GTK+ applications, definitely give it a try!