Prevents situations where the redistributable is installed and OBS
enables the RTX denoiser but it is non-functional. Changes were tested
on systems with both supported / unsupported GPUs and it adds around
10-20ms to the load time in both cases.
Initializing NVAFX is slow as it has to load large models and can cause
significant impact to OBS' startup time. This moves NVAFX initialization
into a separate thread. NVAFX does not appear to be reentrant across
threads, hence a global mutex is also used to avoid reentrancy issues.
The downside of this change is that the first few seconds of audio after
adding the filter or starting OBS will not be filtered, but this is
unlikely to cause any real-world issues.
This minor code cleanup commit does the following:
- sets defaults intensity of RTX denoiser to max.
- adds an initialization check of nvafx.
- splits nvafx initialization from channel allocation for better
readibility of the code.
- moves the intensity update from the filter_audio process function
to the plugin update function.
- logs the error code in case nvafx returns an error when running.
(useful for devs; I haven't translated the error codes which are
available in the sdk).
This fixes issue #4441.
The issue occurs when adding the noise suppression filter for the
first time. Rnnoise or speex are the default noise suppression
methods. Line 344 returns which prevents initialization and
allocation for nvafx (rtx denoiser).
With the fix, initialization of nvafx occurs when swapping methods.
This commit adds support to using Xiph and Mozilla RNNoise library for
noise reduction.
RNNoise is a small library using an AI approach to noise reduction
using a pre-trained model like RTX Voice. But unlike RTX Voice, it is
very tiny, use CPU instead of GPU and only use little resources.
Obviously it is not as efficient but will effectively remove background
noise. It uses more CPU than the existing libspeex-based noise
reduction but it also sounds sounds way better.
RNNoise support is added to the noise reduction effect. It can be
enabled with a checkbox in the effect configuration. RNNoise has no
settings.
Code submissions have continually suffered from formatting
inconsistencies that constantly have to be addressed. Using
clang-format simplifies this by making code formatting more consistent,
and allows automation of the code formatting so that maintainers can
focus more on the code itself instead of code formatting.
Noise Suppression: Clamp sample values before converting to integer.
This fixes an issue where samples exceeding full scale would overflow,
resulting in heavy distortion.
Closesjp9000/obs-studio#1113
(This commit also modifies the following modules: UI,
deps/media-playback, coreaudio-encoder, decklink, linux-alsa,
linux-pulseaudio, mac-capture, obs-ffmpeg, obs-filters, obs-libfdk,
obs-outputs, win-dshow, and win-wasapi)
Adds surround sound audio support to the core, core plugins, and user
interface.
Compatible streaming services: Twitch, FB 360 live
Compatible protocols: rtmp / mpeg-ts tcp udp
Compatible file formats: mkv mp4 ts (others untested)
Compatible codecs: ffmpeg aac, fdk_aac, CoreAudio aac,
opus, vorbis, pcm (others untested).
Tested streaming servers: wowza, nginx
HLS, mpeg-dash : surround passthrough
Html5 players tested with live surround:
videojs, mediaelement, viblast (hls+dash), hls.js
Decklink: on win32, swap channels order for 5.1 7.1
(due to different channel mapping on wav, mpeg, ffmpeg)
Audio filters: surround working.
Monitoring: surround working (win macOs linux (pulse-audio)).
VST: stereo plugins keep in general only the first two channels.
surround plugins should work (e.g. mcfx does).
OS: win, macOs, linux (alsa, pulse-audio).
Misc: larger audio bitrates unlocked to accommodate more channels
NB: mf-aac only supports mono and stereo + 5.1 on win 10
(not implemented due to lack of usefulness)
Closesjp9000/obs-studio#968
These comments have been added to clean up the code and make it more
clear of what the code is doing. The code felt a bit messy, and this
should help prevent the original author of the noise suppression filter
from being lost in case he decides to modify/improve the filter.
When buffering audio data, we don't want to buffer audio data that may
be old. If the audio timing jumps significant and old audio data is
buffered, clear that old data.
The noise suppression filter mistakenly operated on the assumption that
input audio data would always be in 10ms segments, and would crash if
audio data was larger than that size.
Because speexdsp operates on fixed audio frame sizes only, we must
buffer audio data to fit that frame processing size. This creates a
troublesome situation where you must buffer around that specified frame
size.
The new steps for processing are:
1. Push audio data to input circular buffer.
2. Push number of audio frames and timestamp for that audio packet to an
'info' circular buffer.
3. Check size of input circular buffer, and while it's equal to or above
the speexdsp frame size (10ms for minimum latency), pop from the
input buffer to a temporary buffer (10ms frames) and process it, then
push that temporary buffer to the output circular buffer.
4. Peek at the front of the 'info' circular buffer.
5. If the output circular buffer frame size is equal or larger than next
expected number of frames, pop both the info and output buffer, and
return the audio data with the expected audio frames/timestamp.
