Real-time stream mixer with MCU functions¶

Since build 5.2.607 real-time stream mixer function is added. This function with MCU support is designed for real-time video confereincing. Unlike the previous implementation, a real-time mixer does not stop the output stream when some of incoming streams is late (usually due to packet loss or channel interference), and does not wait for bad quality stream to restore.

Configuration¶

Real-time mixer is enabled by default

mixer_realtime=true

Mixer automatic creation when publishing a stream named like user1#room1 is also enabled by default

mixer_auto_start=true

For conferencing, MCU support should be enabled

mixer_mcu_audio=true
mixer_mcu_video=true

It is recommended to reduce mixer idle timeout

mixer_idle_timeout=10000

Stream name displaying in stream picture caption can be enabled if necessary

mixer_display_stream_name=true

in this case caption will be displayed in picture bottom left corner for video streams and in picture center for audio only streams.

Speech indicator greenframe is enabled by default

mixer_voice_activity=true

Another mixer settings are also supported. Note that it is not recommended to enable custom losless video processor because real time function will not work in this case.

Captions management¶

Basic parameters and font size¶

Mixer captions font size can be changed if necessary depending on mixer output stream resolution:

for video streams, caption font size if 20pt by default
```
mixer_font_size=20
```
for audio only streams, caption font size if 40pt by default
```
mixer_font_size_audio_only=40
```

Caption text is also displayed on black rectangle background. The following caption background parameters can be tuned:

Parameter	Default value in pixels	Description
`mixer_text_cut_top`	3	Text top cut
`mixer_text_padding_bottom`	5	Text bottom padding
`mixer_text_padding_left`	5	Text left padding
`mixer_text_padding_right`	4	Text right padding
`mixer_text_padding_top`	5	Text top padding

Text colour, background colour and opacity¶

Since build 5.2.741 caption text colour and background colour can be changed with the following parameters

mixer_text_colour=0xFFFF00
mixer_text_background_colour=0x006666

A colour can be set as hexadecimal value with # or 0x prefix, in #RRGGBB form. Using the settings above, yellow text will be displayed on cian background:

For audio only participants, background colour fills the rectangle.

Since build 5.2.770 caption background opacity can be changed using the following parameter

mixer_text_background_opacity=100

Background opacity should be set in percents: 0 for full transparency, 100 for full opacity (text background will be coloured to defined colour). The default value is 100 (fully opaque).

Font¶

Since build 5.2.770 it is possible to set captions font using the following parameter

mixer_text_font=Serif

The default value is Serif. Font can be chosen from X11 fonts list only, for example:

[root@centos76 ~]# fc-list | grep X11
/usr/share/X11/fonts/Type1/c0611bt_.pfb: Courier 10 Pitch:style=Bold Italic
/usr/share/X11/fonts/Type1/UTBI____.pfa: Utopia:style=Bold Italic
/usr/share/X11/fonts/Type1/c0419bt_.pfb: Courier 10 Pitch:style=Regular
/usr/share/X11/fonts/Type1/c0648bt_.pfb: Bitstream Charter:style=Regular
/usr/share/X11/fonts/Type1/cursor.pfa: Cursor:style=Regular
/usr/share/X11/fonts/Type1/UTB_____.pfa: Utopia:style=Bold
/usr/share/X11/fonts/Type1/c0583bt_.pfb: Courier 10 Pitch:style=Bold
/usr/share/X11/fonts/Type1/UTI_____.pfa: Utopia:style=Italic
/usr/share/X11/fonts/Type1/c0582bt_.pfb: Courier 10 Pitch:style=Italic
/usr/share/X11/fonts/Type1/c0633bt_.pfb: Bitstream Charter:style=Bold Italic
/usr/share/X11/fonts/Type1/c0649bt_.pfb: Bitstream Charter:style=Italic
/usr/share/X11/fonts/Type1/c0632bt_.pfb: Bitstream Charter:style=Bold
/usr/share/X11/fonts/Type1/UTRG____.pfa: Utopia:style=Regular

A full font name should be set, for example

mixer_text_font=Courier 10 Pitch

The default system font will be used if there is no the chosen font in the system.

If the chosen font contains no characters which are in caption text, those characters will not be displayed.

Caption text autoscaling¶

Since build 5.2.709 caption text can be autoscaled to fit the picture. The feature is enabled by default with the following parameter

mixer_text_autoscale=true

If one or more publishers share a screen, a special desktop layout is enabled, in this case desktop stream caption is scaled separately

mixer_autoscale_desktop=true

A minimal font size for text scaling is set by the following parameter

mixer_minimal_font_size=1

to 1pt by default

Room (mixer) name displaying¶

Since build 5.2.770 it is possible to disable room name displaying (which is set to stream name while automatically adding stream to mixer). This can be done with the following parameter

mixer_text_display_room=false

In this case, for stream named user1#room1 only user1 will be displayed (without room name and special character).

Custom stream label displaying¶

Since build 5.2.816 it is possible to display custom stream label in mixer instead of stream and room name. In fact, almost any name can be assigned including national characters.

A label should be set in stream name while publishing, for example

session.createStream({
    streamName: "test1#m1?label=John Doe",
    display: localDisplay,
    ...
}).publish();

In this case the label will be displayed in mixer, including audio only streams

National characters are supported along with the following special characters

~!@#№$%^*()-_,.;:[]{}<>/|\

Character + will be replaced by whitespace, i.e. in the example above stream can be published with label

test1#m1?label=John+Doe

Setting or changing stream label using REST API¶

Since build 5.2.1635 it is possible to set a custom stream label while adding a stream to mixer with /mixer/add REST API query

{
 "uri": "mixer://mixer1",
 "remoteStreamName": "stream1",
 "streamLabel": "John Doe"
}

The label may be changed for the stream already added to the mixer by /mixer/set_stream_label query

{
 "uri":"mixer://m1",
 "remoteStreamName":"stream1",
 "streamLabel": "Mr. John Doe"
}

or

{
 "uri":"mixer://m1",
 "remoteMediaSessionId":"95bf2be8-f459-4f62-9a7f-c588f33e0ad3",
 "streamLabel": "Mr. John Doe"
}

The label may be cleared by passing an empty string as streamLabel

{
 "uri":"mixer://m1",
 "remoteMediaSessionId":"95bf2be8-f459-4f62-9a7f-c588f33e0ad3",
 "streamLabel": ""
}

In this case the stream name will be displayed as published.

Caption location¶

Obsoleted settings in builds from 5.2.844 to 5.2.1077¶

Warning

This setting is not supported since build 5.2.1079!

Since build 5.2.844, it is possible to change location of video participant's caption. By default, caption is located in left bottom corner of video.

It is possible to display caption above video using the following parameter

mixer_text_outside_frame=TOP

or under video

mixer_text_outside_frame=BOTTOM

The distance between video and its caption can be changed using the following parameter

mixer_text_outside_frame_padding=70

Captions placement above or below a picture affects picture placement even if captions diplaying is disabled.

Actual settings since build 5.2.1079¶

Since build 5.2.1079 the following parameter was added to set caption text alignment realtive to frame picture

mixer_text_align=BOTTOM_LEFT

Possible values

Value	Caption placement
`TOP_LEFT`	Top left corner
`TOP_CENTER`	Top center
`TOP_RIGHT`	Top rightcorner
`CENTER`	In a picture center
`BOTTOM_LEFT`	Bottom left corner (by default)
`BOTTOM_CENTER`	Bottom center
`BOTTOM_RIGHT`	Bottom right corner
`EXTERNAL_TOP_CENTER`	Centered above a picture
`EXTERNAL_BOTTOM_CENTER`	Centered below a picture

By default, captions are placed in bottom left corner of a picture frame.

The values EXTERNAL_TOP_CENTER and EXTERNAL_BOTTOM_CENTER replace obsoleted mixer_text_outside_frame parameter. For example, the following value

mixer_text_align=EXTERNAL_BOTTOM_CENTER

will act like mixer_text_outside_frame=BOTTOM

Speech indicator and audio only streams management¶

Thickness and position¶

Speech indicator frame thickness can be set (6 pixels by default)

mixer_voice_activity_frame_thickness=6

The frame is displayed outside astream picture by default. However, if stream pictures are too close, foe example, if CenterNoPaddingGridLayout is used, the frame can affect neighbor pictures

In this case, inner speech frame displaying should be enabled

mixer_voice_activity_frame_position_inner=true

Audio only streams displaying¶

By default, if audio only track from some stream was added to mixer, this stream is shown in separate frame with speech indicator (see above). Sound only can be added to mixer without showing a separate participant (for example, to comment or to dub) with the following parameter if necessary

mixer_show_separate_audio_frame=false

Before build 5.2.965, the stream for dubbing must contain audio and video, and must be added to mixer with the following /mixer/add REST query

{
    "uri": "mixer://mixer1",
    "remoteStreamName": "stream_dub",
    "hasVideo":false,
    "hasAudio":true
}

Since build 5.2.965, audio only stream also will not be displayed if this parameter is used

Since build 5.2.1359, the parameter mixer_show_separate_audio_frame affects stream in mixer displaying when audio data are received before video. In this case, a frame with speech indicator will be displayed for such stream when default settings are used, then video will be displayed after first video data receiving. If the following parameter is set

mixer_show_separate_audio_frame=false

then speech indicator frame will not be displayed until video data arrive.

Avatar pictures displaying for audio only streams¶

Warning

In the builds 5.2.1710 - 5.2.1725 the feature is supported in Ubuntu 20.04 and other systems with glibc 2.31 and newer only

Since build 5.2.1727 the feature is supported in Centos 7.6 (glibc 2.17) and above

Since build 5.2.1710 it is possible to set an avatar picture in PNG (with transparency), JPG or BMP formats to an audio only stream in mixer. A picture may be set while adding a stream to mixer by REST API query /mixer/add:

{
    "uri": "mixer://mixer1",
    "remoteStreamName": "user2",
    "hasVideo": false,
    "hasAudio": true,
    "avatar": "https://mystorage.com/storage/avatar.png"
}

or by REST API query /mixer/set_stream_avatar for the audio only stream which already has added to the mixer:

{
    "uri": "mixer://mixer1",
    "remoteStreamName": "user2",
    "avatar": "https://mystorage.com/storage/avatar.png"
}

The following ways to set a picture are supported:

HTTP URL: https://mystorage.com/storage/avatar.png
file URI: file:///opt/avatar.png
local file: /opt/avatar.png

Audio only stream in mixer is displaying as follows

If stream names displaying in mixer is enabled, audio only stream name will be displayed like video stream one.

Avatar picture may be removed by REST API query /mixer/remove_stream_avatar:

{
    "uri": "mixer://mixer1",
    "remoteStreamName": "user2"
}

Since WCS build 5.2.1858 it is possible to set avatar pictures automatically by stream name, not using REST API. In this case avatar picture files should be placed to the folder

avatar_dir=/usr/local/FlashphonerWebCallServer/avatar

A file name should be equal to stream name. For example, the picture file stream1.png will be applied to the stream named stream1. If the stream was added to the mixer automatically, e.g. stream1#mixer1, the mixer name will not be used to get the picture, only the stream name before # character.

Attention

Avatar picture setting with REST API query /mixer/set_stream_avatar has a higher priority then automatic picture setting

Frame colour¶

Since build 5.2.741 speech indicator frame colour can be changed with the following parameter

mixer_voice_activity_colour=#FF0000

A colour can be set as hexadecimal value with # or 0x prefix, in #RRGGBB form. Using the settings above, speech indicator frame will be red.

Frame displaying when silence is in a stream¶

By default, speech indicator frame may blink at normal participant speech rate, which can be incomfortable for viewing. For this reason, since build 5.2.775 it is possible to set a time interval in which the frame will be displayed since silence is detected in the stream. The feature can be enabled using the following parameter

mixer_voice_activity_switch_delay=500

In this case, speech indicator frame will still be displayed in 500 milliseconds since silence is detected in the stream.

By default, this interval is set to 0, frame will be hidden without delay.

Mixer background management¶

By default, mixer applies black background. Since build 5.2.645 it is possible to set PNG picture file which will be used as mixer background. For example, to change the background to blue, prepare blue.png picture and set the following parameter

mixer_video_background_filename=blue.png

Background picture file should be placed to /usr/local/FlashphonerWebCallServer/conf folder if full path is not set. The file can be placed to any folder on the server, in this case full path to the file should be set to the parameter

mixer_video_background_filename=/opt/media/blue.png

Background picture will be automatically scaled to mixer output stream resolution

Audio only streams picture aspect ratio management¶

By default, audio only streams picture aspect ratio is set according to mixer one (16:9)

mixer_audio_only_width=640
mixer_audio_only_height=360
mixer_video_width=1280
mixer_video_height=720

It can be changed together with mixer if necessary, for example to 4:3

mixer_audio_only_width=640
mixer_audio_only_height=480
mixer_video_width=1280
mixer_video_height=960

Note that audio only streams picture aspect ratio settings are applied only with WCS restart.

Multithreading support and hign load optimizations¶

Since build 5.2.793 multithreading support is added to optimize a mixer for a big participants amount. This feature can be enabled by the following parameters

mixer_type=MULTI_THREADED_NATIVE
mixer_mcu_multithreaded_mix=true

CPU threads count to mix audio amd video can be adjusted by the following parameters

mixer_audio_threads=10
mixer_video_threads=4

Usually, MCU mixer encodes one video stream and a many ausio streams: two per every participant plus one common audio track. Therefore, it is recommended to set more CPU threads for audio encoding than for video (as shown above). Additionally, if freezes occur in MCU mixer output stream, it is recommended to enable multithreaded delivery of mixing result to the main WCS engine, to send it to subscribers.

mixer_mcu_multithreaded_delivery=true

If participants count is lower than mixing threads count (for example, 3 participants), only one CPU thread will be used for mixing.

Real-time mixer tuning¶

Real-time mixer can be tuned using the following parameters

Parameter	Default value	Description
`mixer_audio_silence_threshold`	-50.00	Incoming stream audio silence level in Db
`mixer_debug_mode`	false	Adds some debug info to stream picture caption
`mixer_in_buffering_ms`	200	Incoming stream video buffer in milliseconds
`mixer_incoming_time_rate_lower_threshold`	0.95	Relative incoming stream time to mixer time rate lower threshold
`mixer_incoming_time_rate_upper_threshold`	1.05	Relative incoming stream time to mixer time rate upper threshold
`mixer_video_stable_fps_threshold`	15	Incoming stream FPS threshold, video buffer will be disabled for streams with low fps
`audio_mixer_max_delay`	300	A maximum time interval to wait for any audio data in the incoming stream, in milliseconds, when this time is expired, the stream will not be mixed until some media data receiving

Testing¶

For test we use
- demo server demo.flashphoner.com;
- Chrome browser;
- MCU Client web application example for conferencing.
Open MCU Client web application page. Enter user name user1 and room name room1
Click Join. A stream from your web camera will be published and added to mixer, then mixer output stream will be displayed without your microphone audio
Open MCU Client application page in another brower tab/browser/PC. Enter user name user2 and room name room1
Click Join. A stream from your web camera will be published and added to mixer, then mixer output stream will be displayed with videos from both users but audio from user1 only
On user1 page two videos also are playing with user2 audio only
Click Leave to leave the room in both tabs/browsers

Call flow¶

Call flow for conferencing example based on real-time mixer with MCU function is described on MCU Client page.

Incoming streams tuning recommendations¶

When delay occurs in the incoming stream from one of participants, realtime mixer will freeze that stream. The following is recommended to minimize incoming stream delays:

For RTMP streams, adjust encoding parameter so that:
- client encoder perfomance to be enough to send picture frames in time
- stream resolution and bitrate to fit to publishers channel from client to server
For WebRTC streams, do not raise minimum video bitrate threshold higher than webrtc_cc_min_bitrate server configuration parameter defines. By default, lower video bitrate threshold is set to 30 kbps
```
webrtc_cc_min_bitrate=30000
```
The publisher client browser will adopt the stream to channel quality drops. The lower bitrate the lower picture quality, but the participant stream will not be freezed in this case.

Adding one stream to two or more realtime mixers simultaneously¶

Since build 5.2.732 one stream can be added to two ore more realtime mixers simultaneously. Note that realtime mixer should be enabled

mixer_realtime=true

and custom losless videoprocessor should be disabled

mixer_lossless_video_processor_enabled=false

Known issues¶

1. Real-time mixer functions are disabled when custom losless videoprocessor is used¶

Real-time mixer functions are disabled when custom losless videoprocessor is used, mixer incoming streams cannot be played in this case.

Symptoms

Stream stops playing after adding to mixer

Solution

Do not use custom losless videoprocessor with real-time mixer

mixer_lossless_video_processor_enabled=false

`fontconfig` library must be installed on server to display stream caption¶

To display stream captions in mixer, it would be necessary to install fontconfig library

Symptoms

Streams cannot be added to mixer with the following exception in server log

09:17:11,756 ERROR           MixerAgent - MIXER-AGENT-mixer://mixervmixdr52-9d46cd04-5867-4d74-a9d9-baf67f74e7d2 Mixer closed due to error
java.lang.InternalError: java.lang.reflect.InvocationTargetException
   at java.desktop/sun.font.FontManagerFactory$1.run(FontManagerFactory.java:86)
   at java.base/java.security.AccessController.doPrivileged(AccessController.java:310)
   at java.desktop/sun.font.FontManagerFactory.getInstance(FontManagerFactory.java:74)
   at java.desktop/sun.font.SunFontManager.getInstance(SunFontManager.java:247)
   at java.desktop/sun.font.FontDesignMetrics.getMetrics(FontDesignMetrics.java:265)
   at java.desktop/java.awt.Font.getStringBounds(Font.java:2606)
   at java.desktop/java.awt.Font.getStringBounds(Font.java:2516)
   at com.flashphoner.media.N.A.A.A(Unknown Source)
   at com.flashphoner.media.mixer.video.presentation.Canvas.computeTextScales(Unknown Source)
   at com.flashphoner.media.mixer.video.presentation.Canvas.writeNative(Unknown Source)
   at com.flashphoner.media.N.A.A(Unknown Source)
   at com.flashphoner.media.N.D.D(Unknown Source)
   at com.flashphoner.media.N.D.A(Unknown Source)
   at com.flashphoner.server.remote.B.B.Г™(Unknown Source)
   at com.flashphoner.server.remote.B.B.run(Unknown Source)
Caused by: java.lang.reflect.InvocationTargetException
   at java.base/jdk.internal.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method)
   at java.base/jdk.internal.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:62)
   at java.base/jdk.internal.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:45)
   at java.base/java.lang.reflect.Constructor.newInstanceWithCaller(Constructor.java:500)
   at java.base/java.lang.reflect.Constructor.newInstance(Constructor.java:481)
   at java.desktop/sun.font.FontManagerFactory$1.run(FontManagerFactory.java:84)
   ... 14 more
Caused by: java.lang.NullPointerException
   at java.desktop/sun.awt.FontConfiguration.getVersion(FontConfiguration.java:1262)
   at java.desktop/sun.awt.FontConfiguration.readFontConfigFile(FontConfiguration.java:225)
   at java.desktop/sun.awt.FontConfiguration.init(FontConfiguration.java:107)
   at java.desktop/sun.awt.X11FontManager.createFontConfiguration(X11FontManager.java:719)
   at java.desktop/sun.font.SunFontManager$2.run(SunFontManager.java:367)
   at java.base/java.security.AccessController.doPrivileged(AccessController.java:310)
   at java.desktop/sun.font.SunFontManager.<init>(SunFontManager.java:312)
   at java.desktop/sun.awt.FcFontManager.<init>(FcFontManager.java:35)
   at java.desktop/sun.awt.X11FontManager.<init>(X11FontManager.java:56)
   ... 20 more

Solution

Install fontconfig library:

sudo yum install -y fontconfig

3. A participant name can be drawn over speech indicator frame if stream picture is small enough¶

A participant name can be drawn over speech indicator frame if there are many participants in mixer and therefore participants picture is small

Symptoms

For small participant pictures (in desktop layout, for example) participants name can be drawn over speech indicator frame

Solution

Enable inner speech indicator frame position and decrease frame thickness

mixer_voice_activity_frame_position_inner=true
mixer_voice_activity_frame_thickness=2

4. Mixer stream can play unsmoothly when participants stream contains B-frames¶

B-frames should be excluded from mixer input stream, or mixer input buffer should be increased

Symptoms

When mixer input stream is encoded by Main profile with B-frames, the mixer output stream is playing unsmoothly, there are short freezes with sound pauses

Solution

a) encode stream without B-frames (preferrable)

b) increase mixer input buffer (may give an additional delay)

mixer_in_buffering_ms=600

5. An incoming stream with a high FPS may give additional server load¶

60 FPS stream publishing to mixer gives additional CPU and system memory load

Symptoms

CPU load increasing and system memory (not Java heap) consumption growing while publishing 60 FPS stream to a mixer

Solution

Use more powerful server or publish 30 FPS and lower streams to mixer

6. Captions may twitch for many participants in MCU mixer¶

Symptoms

For many (over 10) participants in MCU mixer captions may periodically twitch

Solution

Apply the following settings

mixer_text_bulk_write=false
mixer_text_bulk_write_with_buffer=false

7. Some participants are less audible then others when speaking simultaneously¶

When a number of participants are speaking simultaneously, some participants are less audible then others. The problem is typical for mixing WebRTC streams, and may ocuur in any solutions using WebRTC and audio mixing, in Discord for example: more data are encoded to one stream, more audio samples may be dropped due to fixed samplerate.

Symptoms

When three or more participants are speaking simultaneously, one of them is less audible then two others

Solution

Raise the audio publishing bitrate on client side to send more data per one participant

constraints: {
  audio: {
    bitrate: 128000
  }
}

and raise Opus encoding bitrate on server

opus.encoder.bitrate=128000

8. An incoming stream will be never encoded in realtime mixer if media traffic is stopped in the stream¶

Symptoms

An incoming stream freezes in the mixer when media traffic is stopped in the stream, for example, an application window captured by screen sharing is minimized to task bar

Solution

Decrease a minimal FPS threshold for mixer incoming streams

mixer_video_stable_fps_threshold=0

9. Mixer output stream bitrate may be unstable when incoming stream bitrate or fps drops¶

Symptoms

Mixer output stream bitrate becomes unstable when incoming stream bitrate or fps drops

Solution

Update WCS to build 5.2.1843 or newer and set the parameter

h264_encoder_filler_data_padding=true

10. There may be a stuttering in mixer audio for a stream published with a packets loss¶

Symptoms

Micro freezes and stuttering occur for a certain stream in mixer

Solution

Increase the mixer incoming streams buffer

mixer_in_buffering_ms=1000

Real-time stream mixer with MCU functions¶

Configuration¶

Captions management¶

Basic parameters and font size¶

Text colour, background colour and opacity¶

Font¶

Caption text autoscaling¶

Room (mixer) name displaying¶

Custom stream label displaying¶

Setting or changing stream label using REST API¶

Caption location¶

Obsoleted settings in builds from 5.2.844 to 5.2.1077¶

Actual settings since build 5.2.1079¶

Speech indicator and audio only streams management¶

Thickness and position¶

Audio only streams displaying¶

Avatar pictures displaying for audio only streams¶

Frame colour¶

Frame displaying when silence is in a stream¶

Mixer background management¶

Audio only streams picture aspect ratio management¶

Multithreading support and hign load optimizations¶

Real-time mixer tuning¶

Testing¶

Call flow¶

Incoming streams tuning recommendations¶

Adding one stream to two or more realtime mixers simultaneously¶

Known issues¶

1. Real-time mixer functions are disabled when custom losless videoprocessor is used¶

fontconfig library must be installed on server to display stream caption¶

3. A participant name can be drawn over speech indicator frame if stream picture is small enough¶

4. Mixer stream can play unsmoothly when participants stream contains B-frames¶

5. An incoming stream with a high FPS may give additional server load¶

6. Captions may twitch for many participants in MCU mixer¶

7. Some participants are less audible then others when speaking simultaneously¶

8. An incoming stream will be never encoded in realtime mixer if media traffic is stopped in the stream¶

9. Mixer output stream bitrate may be unstable when incoming stream bitrate or fps drops¶

10. There may be a stuttering in mixer audio for a stream published with a packets loss¶

`fontconfig` library must be installed on server to display stream caption¶