Exporting produces artifact-heavy files

Hi guys, I'm on Linux and using LMMS 1.1.3 from the Debian `testing` release. I am absolutely sure that I have been able to export files normally a few months ago so I'm thinking this has been caused by updates in related libraries and not in the program itself - since LMMS hasn't been updated in a while, as far as public releases go. Anyway there seems to be a part of this that could be internal so I'm following with the report.

My problem is that while the song plays fine inside LMMS, when exporting it it results in a very artifact-heavy (noisy) file, which is unusable for any practical purposes. I'm using the default export settings and both OGG and WAV formats produce the same results.

What makes this interesting is that when I open LMMS, it will produce the same type of noise while playing, but as soon as I change the audio backend from Pulseaudio to ALSA (or SDL) the issue is solved and I can use LMMS normally. However, after an export, the audio backend seems to be set to Pulseaudio again automatically - which is why I'm betting that this is the underlying issue (exporting with Pulseaudio instead of chosen backend, if that makes any sense).

After an export and the noisy file being generated, the application will once again be very noisy until I set the backend back again to ALSA or SDL. Is it possible this "broken" backend is being used to export my song, producing a broken file? In the least, I think, the exporting process should not change my backend preferences - or at least return it to my previous choice after the exporting is done with.

Some further package version information for my system is going below, maybe it will prove useful. Thank you all very much for the work on LMMS - I can't imagine ever using another composing software, regardless of what type of work I'm doing, platform I may be using or budget for my project :) You guys rock! (btw I'm pretty excited for the next release, coming with MIDI exporting too!)

```
Kernel: Linux 4.9.0-1-amd64

Versions of packages lmms depends on:
ii calf-ladspa 1.1.3-7
ii libasound2 1.1.3-4
ii libc6 2.24-9
ii libfftw3-single3 3.3.5-3
ii libfltk1.3 1.3.4-2
ii libfluidsynth1 1.1.6-4
ii libgcc1 1:6.3.0-6
ii libjack-jackd2-0 [libjack-0.125] 1.9.10+20150825git1ed50c92~dfsg-4+b1
ii libogg0 1.3.2-1
ii libportaudio2 19.6.0-1
ii libpulse0 10.0-1
ii libqt4-xml 4:4.8.7+dfsg-11
ii libqtcore4 4:4.8.7+dfsg-11
ii libqtgui4 4:4.8.7+dfsg-11
ii libsamplerate0 0.1.8-8
ii libsdl1.2debian 1.2.15+dfsg1-4
ii libsndfile1 1.0.27-1
ii libstdc++6 6.3.0-6
ii libstk-4.5.0 4.5.2+dfsg-4
ii libvorbis0a 1.3.5-4
ii libvorbisenc2 1.3.5-4
ii libvorbisfile3 1.3.5-4
ii lmms-common 1.1.3-7
ii stk 4.5.2+dfsg-4
ii zlib1g 1:1.2.8.dfsg-5

Versions of packages lmms recommends:
ii caps 0.9.24-4
pn lmms-vst-server:i386
ii tap-plugins 0.7.3-2

Versions of packages lmms suggests:
ii calf-ladspa [ladspa-plugin] 1.1.3-7
ii caps [ladspa-plugin] 0.9.24-4
ii fil-plugins [ladspa-plugin] 0.3.0-6
ii fluid-soundfont-gm 3.1-5.1
ii freepats 20060219-1
ii mcp-plugins [ladspa-plugin] 0.4.0-5
ii omins [ladspa-plugin] 0.2.0-7.1
ii swh-plugins [ladspa-plugin] 0.4.17-1
ii tap-plugins [ladspa-plugin] 0.7.3-2
```

No source files or tests are named. Start by reproducing export with PulseAudio, ALSA, and SDL using both OGG and WAV, then inspect how exporting selects and restores the audio backend; done means exported files are free of artifacts and the user's selected backend is preserved.

Written by the indexing model from the issue text.