Extending The Effects Engine Report W7

Hello,

This week I started with fixing some bugs on the branch[1] responsible for replacing the default equalizers with a dedicated EffectRack. One of them was causing Mixxx not to shutdown properly, it wasn't getting all the way down to return 0 line in main.cpp. The problem was that inside DlgPrefEQ's destructor I was deleting the data from m_pConfig, which is a pointer to Mixxx's configurations. As you'd expect, the same data was being deleted inside MixxxMainWindow because that's where it was initialized in the first place. This double deletion of the same data leaded to abnormal termination of the program.

My next task was displaying human readable names inside the QComboBoxes for selecting which effect is active for a specific deck. Each item from the QComboBox has an additional data field which I used to store the effect ID. The Preference page has a check box which switches between showing all the effects or only the equalizer ones. To avoid coupling between DlgPrefEq and effects I introduced a flag for each effect which is set to true if the effect is an equalizer. Inside EffectsManager there are separate methods for retrieving either all the available effects or only the equalizers which are used to populate the drop down lists. To set the name and data for each entry I used two separate lists: one for names and one for IDs. I wrongly assumed that each effect had the same index in both lists because the routine for obtaining them was similar. This was not true because a QSet was involved in the process. I did some research and a QSet is implemented with a QHash which does not guarantee that its items are stored in the same order they are introduced. I fixed this issue by storing both the name and the ID of an effect inside the same QSet.

The second part of my GSoC project is adding LV2 support to Mixxx. I started with testing Calf[2] plug inside Audacity and Ardour. Audacity 2.0.5 which is the last released version does not support them, so I had to compile Audacity 2.0.6 from sources. After installing and uninstalling libraries, resolving conflicts between packages I managed to get it up and running. Afterwards I found two Jack hosts for LV2: calfjackhost (comes with calf plug ins) and jalv (written by the creator of LV2). I installed jack, qjackctl and after fixing an issue with jack server not being able to start, I was able to route Mixxx's sound through LV2 plug ins and then back to speakers. I tested jack on Windows too and it seems to work as expected which is good news for us. This made me and Daniel consider another approach to adding LV2 support to Mixxx. Rather than making Mixxx an LV2 host, use an existing one to process the sound. Roughly, Mixxx will feature an effect which routes the sound samples to a Jack host, get them back modified and output them to Mixxx's master output. This approach has the advantage of being more stable and less error prone.

As a side task I took on the quest to implement a FFT Graphic Equalizer for Mixxx, based on Audacity's Equalization[3]. After days of reading about the Fourier transform, FFT windowing[4] and overlapping[5], I was disappointed to conclude that is not worth having such an Equalizer. Audacity has a window size of ~16000 samples and it is filtering samples in chunks to use the overlap add method. However, since Mixxx is doing real time equalization, we don't have access to many samples (I did some tests and I got 8192 samples when I chose the largest audio buffer from preferences). This implied to set the window size to our buffer size and further divide this buffer into chunks. Consequently we applied FFT on a small number of samples which gave us poor frequency range (FFT_sample_len / 2 frequency bins were accessible). This[6] post gives some arguments against frequency domain equalizers. As we were aiming for an efficient equalizer, here is a quote from that blog post which is proving the contrary: The FFT, though efficient compared to the DFT (which is the FFT without the "fast" part), performs worse than linear time, and we need to do both the FFT and it's inverse, which is computationally similar. EQing with the FFT is therefore generally very inefficient compared to comparable time-domain filters.

I was confused by the terms PortAudio, PulseAudio, Jack, ALSA and so on. Playing with jack, Mixxx preferences and talking with a few guys on IRC made me learn new and interesting things about Linux sound architecture. I found out that PortAudio is a cross-platform library for controlling various lower level components such as PulseAudio, Jack, Alsa, etc. PulseAudio is similar with Jack. They do their job through ALSA or another low level piece of software like the older OSS.

Yours truly,
Nicu Badescu

[1] - https:*github.com/badescunicu/mixxx/tree/kill_buttons
[2] - http:*calf.sourceforge.net/
[3] - https:*github.com/mcpierce/audacity/blob/master/src/effects/Equalization.cpp
[4] - http:*www.physik.uni-wuerzburg.de/\~praktiku/Anleitung/Fremde/ANO14.pdf
[5] - http:*www.zytrax.com/tech/audio/equalization.html\#fft
[6] - http:*blog.bjornroche.com/2012/08/why-eq-is-done-in-time-domain.html

Back to the main page!