Capture monitor devices with navigator.mediaDevices.getUsermedia() at Chromium and Chrome on Linux

capture_monitor_devices_at_chromium_and_chrome_on_linux.md

Chromium and Chrome refus to list or capture monitor devices on Linux

• Issue 931749: DOMException: could not start audio source when trying to access audioinput
• Issue 1155954: Add --allow-monitor-inputs flag
• Issue 865799 in chromium: A way for enumerateDevices() to show audiooutput devices with labels when user has no cam or mic

Virtual microphone using GStreamer and PulseAudio includes a section that describes a workaround using PulseAudio to create a virtual microphone where the source is a monitor device

Remap source

While the null sink automatically includes a "monitor" source, many programs know to exclude monitors when listing microphones. To work around that, the module-remap-source module lets us clone that source to another one not labeled as being a monitor:

pactl load-module module-remap-source \ master=virtmic.monitor source_name=virtmic \ source_properties=device.description=Virtual_Microphone

After we run

$ pactl load-module module-remap-source master="$(pactl list | grep -A2 '^Source #' |  grep 'Name: .*\.monitor$' | awk '{print $NF}' | tail -n1)" source_name=virtmic source_properties=device.description=Virtual_Microphone

we can call navigator.mediaDevices.getUserMedia({audio: true}) to get permission to access device labels, filter the devices from navigator.mediaDevices.enumerateDevices() to get the device 'Virtual_Microphone' then call navigator.mediaDevices.getUserMedia({audio: {deviceId: {exact: deviceId}}}) again, deviceId is the deviceId of 'Virtual_Microphone' with the source of theat virtual microphone set to the monitor device that we want to capture.

I've been looking everywhere for this, thank you

@benwiley4000 Note, during experiments the quality of audio from getUserMedia() is sub-par compared to processing raw PCM https://github.com/guest271314/captureSystemAudio/tree/master/native_messaging/capture_system_audio. I suggest trying both approaches yourself.

@guest271314 did you try processing raw PCM from getUserMedia using a media stream source/AudioWorklet(or ScriptProcessorNode) with web audio API? That's what I'm doing although I don't know if there's somehow a loss of quality in the middle. I'm not using a MediaRecorder.

BTW what is the purpose of the "exact" field? What happens if I pass the device ID without "exact"?

I can try to compare results with the python script later but in case you're curious here is my app, it lets you record an up-to-10 second clip from any input device and also lets you save the wav recording as a file (which comes from raw PCM): https://benwiley4000.github.io/volca-sampler/

For the source code see:
Main thread: https://github.com/benwiley4000/volca-sampler/blob/master/src/utils/recording.js
AudioWorklet: https://github.com/benwiley4000/volca-sampler/blob/master/public/recorderWorkletProcessor.js

did you try processing raw PCM from getUserMedia using a media stream source/AudioWorklet(or ScriptProcessorNode) with web audio API? That's what I'm doing although I don't know if there's somehow a loss of quality in the middle. I'm not using a MediaRecorder.

Yes. https://github.com/guest271314/AudioWorkletStream.

BTW what is the purpose of the "exact" field? What happens if I pass the device ID without "exact"?

To unambiguously set the exact deviceId https://w3c.github.io/mediacapture-main/#constrainable-properties. Monitor devices are not listed by default on Chromium on Linux. We are sure to not get the 'Default' device, or other microphone.

What is the result of your test?

For the source code see:
Main thread: https://github.com/benwiley4000/volca-sampler/blob/master/src/utils/recording.js

Note, I have found using a Blob to store data to be far faster than using Array.prototype.reduce(), see https://bugs.chromium.org/p/chromium/issues/detail?id=1260519.

I've been setting device ID after initial stream to a different ID with no problem but I guess it wouldn't hurt to change to use the exact property. Also thanks for the tip about Blobs. For now I don't have any noticeable performance issues but I will consider this for the future. Le mar. 2 nov. 2021 01 h 01, guest271314 ***@***.***> a écrit :

…

***@***.**** commented on this gist. ------------------------------ did you try processing raw PCM from getUserMedia using a media stream source/AudioWorklet(or ScriptProcessorNode) with web audio API? That's what I'm doing although I don't know if there's somehow a loss of quality in the middle. I'm not using a MediaRecorder. Yes. https://github.com/guest271314/AudioWorkletStream. BTW what is the purpose of the "exact" field? What happens if I pass the device ID without "exact"? To unambiguously set the exact deviceId https://w3c.github.io/mediacapture-main/#constrainable-properties. Monitor devices are not listed by default on Chromium on Linux. We are sure to not get the 'Default' device, or other microphone. What is the result of your test? For the source code see: Main thread: https://github.com/benwiley4000/volca-sampler/blob/master/src/utils/recording.js Note, I have found using a Blob to store data to be far faster than using Array.prototype.reduce(), see https://bugs.chromium.org/p/chromium/issues/detail?id=1260519. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <https://gist.github.com/53e00c6765aa256362fb52c08e82d189#gistcomment-3948065>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADHOD3P4UVUAMBRSKFTPEZDUJ5WDDANCNFSM5HFGOJOA> .

Thanks for the gist. I have a query. I am using pipewire only now. Don't want to use pulseaudio at all. I found something similar for pipewire called pw-loopback. What should be the command for pipewire?

I have created a virtual sink with pw-loopback -m '[ FL FR]' --capture-props='media.class=Audio/Sink node.description=discord-sink node.name=my-sink'
and virtual source using pw-loopback -m '[ FL FR]' --playback-props='media.class=Audio/Source node.description=discord-source node.name=my-source'.

Now patchbay looks like this. I have selected discord source as mic and discord-sink as output device in discord. I want to stream audio from Lollipop music player to discord(google chrome input). How to route this?

Again thanks a lot for the gist.

@hasrack I have no experience using Discord, see https://github.com/Lightcord/Lightcord/issues/31.