Discord Linux Go-Live Audio Workaround Bot

Good news!

Discord intends to eventually provide native audio sharing via PipeWire as confirmed by one of their staff on reddit.

These workarounds at this point only exist for while we wait.

Projects made by others, that try to deal with this problem:

• someone managed to get Go-Live audio to work by streaming via a browser and applying custom scripts. You can check it out here.
• A custom client that supports streaming with audio

These may be worth a try if you don't want to use my bot.

Backstory

It's been a very very long time since Go Live was rolled out for Linux users of Discord, but as of 2023/03/31, they still haven't fixed the issue where audio is not being captured from the application that is being streamed. You can call this a missing feature if you really want to be technical, but an 'average' end user will disagree.

My bug report was denied on their Discord Testers server
with the following reason

I would have been happy if they at least gave us some sort of workaround, like starting a capture from an automatically created sink and telling us to divert app audio to it, and it will be sent the same way the Go Live audio is sent, but nothing of sort was ever rolled out.

These reasons + trying many other methods I detail at the bottom of this readme and the fact that I am a somewhat experienced Discord bot developer inspired me to make this bot.

So, this is a Discord bot that will allow you to stream application audio through it to a voice channel the exact same way those music bots do.

The advantages of using this rather than using audio routing solutions through microphone input of Discord are that:

1. Discord mic input is encoded in mono, while bots can stream stereo audio. Mono audio sounds terrible, especially if you are streaming a rhythm game like osu!
2. You will need to give up noise suppression etc when doing it through the mic.
3. Allow the end user audio level adjustment of the stream audio and your voice individually.
4. End users who don't want to listen to your stream can just mute the bot.

---

Installation Instructions

1. Pick a .py file from this repo, name it what you want and put it in your PATH if you want to.
   • go-live-bot-classic.py -- uses old style prefixed commands, scraping messages.
   • go-live-bot-slash.py -- uses slash commands. (This one is still work in progress. Use classic.)
2. Make the file executable chmod +x go-live-bot-classic.py
3. Install discord.py 2.x with voice support: pip3 install discord.py==2.2.2
   • You may also need to install PyNaCl using: pip3 install PyNaCl
4. Make sure you have ffmpeg and pavucontrol installed.
5. Set GOLIVE_BOT_TOKEN environment variable with the bot token.

To get a bot token, register a new app here, create a bot, and copy the token (not the client secret).
You also need to enable MESSAGE CONTENT intent for the classic command version and SERVER MEMBERS intent for being able to use the join commands in DMs without specifying a channel ID.

How to use

1. Using the commands bellow, make a virtual audio sink named STREAM_AUDIO

   pacmd load-module module-null-sink sink_name=STREAM_AUDIO
   pacmd update-sink-proplist STREAM_AUDIO device.description=STREAM_AUDIO

   • Note: use pactl instead of pacmd if you are using PipeWire.
   • (optional, experimental) At the top of the file in FFMPEG_PULSEAUDIO_SOURCE, you can put CUSTOM_SINK.monitor and skip step 7.
2. Start the bot. (To start the bot, literally run it.)
3. Send application audio to this sink. If you can't do it through the application, try using pavucontrol for that.
4. Using the following command, create a stereo virtual audio loopback session.

   pactl load-module module-loopback latency_msec=1 channels=2

5. Using pavucontrol make the loopback device take audio from the monitor of our sink we just made.
   • For some reason, in pavucontrol the sink may show up as Output to Null Device.
   • This is, so you can hear the application audio yourself without a delay.
6. Type /join (optionally followed by the voice channel ID in a chat where the bot can read).
   • This will make the bot join and start streaming your microphone.
7. Using pavucontrol, locate the recording session that the bot is doing, and change the input the monitor of your STREAM_AUDIO.
   • For some reason, in pavucontrol the sink may show up as Output to Null Device
8. profit???
9. When you are done, type /leave in a chat where the bot can read.
10. After that, you can just right-click in pavucontrol on the loopback device and click on terminate.
    • Or you could just type pulseaudio -k in the terminal.

---

Known issue + Help wanted:

If you get an error like default: Generic error in an external library, I have no idea what causes this.
You are welcome to find a solution and make a PR or create an issue. I'll investigate when I have time.

---

Note: if you are having problems with audio cutting out, try export PULSE_LATENCY_MSEC=5

---

My adventures of exploring this bug.

I explored this bug greatly, apparently it's a bug in electron from 5+ years ago but it was closed without actually being fixed, presumably waiting for upstream, Chrmium to do something about it. I explored many solutions to this problem, spent countless hours on it and making this bot was the best solution I could come up at the time.

Other solutions I tried, include:

1. Streaming through the web version
   • Tried in Chromium
     • Even the Share audio button does not work when sharing a tab
   • Tried in Firefox
     • Discord normally won't let you, so I had to set the user agent to represent Chromium
       • UPDATE: user agent spoofing is no longer needed
   • Even on Windows, streaming using the web version in Chrome does not capture audio
     • Even when streaming just a tab and checking Share audio button
   • Not sure if this is broken in the web app, or the Chrome/Chromium is broken.
     • If on the web app, perhaps a browser extension can fix this? I'm not good enough to make one
     • UPDATE: this was a bug with discord
   • UPDATE, I reported this bug and after several months, they marked it as fixed, around the time they added this feature to mac.
     • But it applies noise suppression on it. But at the time I reported it, it transported no audio at all.
       • Maybe this will be fixed in few weeks? if not I will just file a new bug report.
         • UPDATE:
         • UPDATE 2: someone made a script that fixes this, see this
     • But having this only limits us to screen-sharing browser tabs with sound. It's better than nothing I guess.
       • Maybe we could use obs/ffmpeg to stream to a browser tab and screen-share a playback of that? we could also locally mute the tab while it's still sharing audio
       • Or maybe someone can convince the Chromium devs to add this for Linux as well. Maybe through PipeWire?
       • I wonder if someone can add this to Chromium via browser extensions? I don't know what their limits are though.
2. Running Windows version of Discord in wine.
   • The screen-share would not work. The DiscordHookHelper.exe would crash. Tried various options relating to Hardware acceleration on/off, didn't help.
   • I imagine many APIs required for this to happen may not be implemented in wine.
3. Dedicated, real Windows environment for running Discord.
   • OBS streaming to a custom RTMP server and screen-sharing a mpv playback of that. Creating a virtual audio output devices and make obs pick up audio from that, so voice loopback wouldn't happen
     • This is painfully slow in a VM
     • This can be done on a dedicated computer through LAN but there is a good 5-second delay which I was not able to reduce
       • Most of the delay is probably happening during the encoding. Maybe we could move to an ffmpeg based solution?
     • For a solution like this, using ffmpeg with a custom rtmp server sounds like the best idea. on the windows end, we screen-share an MPV view.
   • Capturing with a capture card and playing back with mpv or vlc and screen-sharing that.
     • could easily mirror the screen with Xorg.conf
     • Still need to deal with playing back audio, vlc can do both video and audio at the same time
     • Capture cards are expensive, cheap ones have mediocre quality both video and audio wise. the $16 MACROSILICON one has mono audio anyways, which defeats the purpose of all this, mostly.
       • To get around this, we can separately send the audio, whether though LAN or AUX cable. Then we could just stitch it using MPV.
     • In case of a VM, USB-passthough is a thing,
       • Although as previously started, this is painfully slow in a VM. Just playing back a video stream in a VM takes a sizable amount of processing power.
   • Did I mention that attempting to screen-share in VMWare blue-screens the whole VM? Well, unless you disable Hardware Acceleration in Virtual machine settings.
   • I am not sure how well hardware accelerated video playback is a thing in virtual machines?
   • Also, when I mention, virtual machines being slow, I am not talking about PCIe pass-though. that can help out a lot.
4. Using discord_arch_electron package from AUR. this uses system installation of electron instead of what discord bundles.
   • So in an event this bug is fixed in electron, we may get the fix instantly (maybe with a BetterDiscord plugin mixed in?)
     • or if we decide to throw together a hacked version of electron that just makes a sink when appropriate function is called
     • very unlikely, because it may require some other modifications on the web app side
       • as stated, it's possible we can do those with BetterDiscord, see the next section.
   • I'm not good enough to just fix this bug but this is a good starting point if you want to give it a try
     • and maybe give them a PR after fixing, and if they reject it, just make a PKGBUILD repo that applies your patch, put on GitHub, doesn't matter.
     • This is like, the best solution to this problem, if you are good enough.
5. A BetterDiscord plugin
   • I tried this approach as well and got absolutely nowhere, due to the lack of experience I have.
     • There is/was literally zero documentation about the relevant things necessary for this to happen.
   • This could go along with discord_arch_electron in an event this is fixed in electron.
   • Maybe if there is someone who knows how to make BetterDiscord plugins could rig something up?
     • Maybe give us a sink we can direct audio to.
   • Since this was fixed on a Mac, maybe someone could try that 'Emulator' plugin BD has when on canary it gets updated to support the new BD and see where it goes
6. Reverse engineer how Discord sends a stream and make a small console based app to run an encode with and stream.
   • preferably it would just emulate an RTMP server where OBS or ffmpeg would stream to, and it would forward that to discord.
     • although, discord does not stream in h264, it streams in vp9 iirc to save bandwidth. or maybe it depends.
   • No discord bot lib has implemented something like this, so you're on your own
   • having a client that does not behave 100% like discord servers expect it to, may flag your account and get you banned. but hey, what else do we linux users even expect for companies like these?

I'm still very inexperienced, so this is the best I can come up with. For me, adding this to Chromium sounds like the most likely and sustainable thing with the least risk for a ban.