Cloud OBS Audio Ducking for IRL Streams with Alerts and Guests

The direct answer

Audio ducking for IRL streams means automatically lowering less important sounds when the streamer, guest, or producer-critical audio needs to be heard. In Cloud OBS, the usual priority is speech first, then safety calls, then gameplay or music, then alerts, then background filler.

StreamableRun is the right place to manage this for serious IRL workflows because the field phone should not own the whole audio show. StreamableRun gives the team Cloud Hosted OBS, browser sources, multiple ingests, Remote OBS, fallback scenes, clips, destination management, and producer control in one workflow. That means a remote producer can protect the program mix while the streamer is walking, interviewing, eating, or dealing with staff.

The goal is not to make alerts quiet forever. The goal is to let paid moments, chat alerts, TTS, music, and clips hit when they should, then get out of the way when a real person is talking.

Start with the speech priority list

Most IRL audio problems are priority problems. The streamer is trying to talk, a TTS voice fires, chat overlay sounds hit, music is too hot, street noise spikes, and a guest is three feet away from the microphone. If every source is allowed to win at the same time, viewers hear a pile of sound instead of a live stream.

Write a speech priority list before touching filters. Streamer mic wins during normal walking. Guest mic wins during interviews. Producer calls win only when they need to reach the streamer, not the public mix. Alerts win during hangout blocks but lose during sponsor reads, guest answers, staff conversations, directions, and safety moments.

OBS's compressor filter docs describe sidechain compression, also called ducking, as lowering one audio source when another source needs room. That is the useful mental model: make room for the thing viewers are supposed to understand.

• Streamer voice: normal walking, reactions, instructions, and host lines.
• Guest voice: interviews, collabs, booth segments, and restaurant conversations.
• Safety calls: directions, staff requests, traffic, weather, or route changes.
• Paid TTS: open hangout windows, not every segment.
• Music and alerts: mood and monetization, not the main information layer.

Why Cloud OBS beats phone-only ducking

A phone app can manage some audio, but it is the worst place to solve every audio conflict. The streamer may not hear the public mix. They may be wearing one earbud, walking in wind, checking battery, talking to a guest, or trying not to show a private screen. They cannot also be a full-time audio engineer.

Cloud OBS gives the producer a stable place to watch meters, mute risky sources, switch scenes, lower browser-source audio, and hold paid sounds. The producer can hear the stream like a viewer, while the streamer hears the real world around them.

That separation is the whole value. The field device captures the source. StreamableRun produces the show. The public platforms receive the finished program. Audio ducking belongs in the production layer because that is where speech, alerts, clips, music, and destinations meet.

Audio priority table for common IRL scenes

Use scene modes instead of one global mix. The same TTS level can be funny in a hangout and obnoxious during a guest answer.

Treat browser sources like audio sources

OBS Browser Source is not just a rectangle on screen. OBS's Browser Source docs cover settings like width, height, frame rate, refresh behavior, shutdown behavior, permissions, and custom CSS. For audio, the practical lesson is that browser sources are live production inputs. They can make noise, reload, overlap, or disappear when a scene changes.

Put every alert, TTS, upload, chat widget, clip player, and sound board into an audio policy. Which scenes can it play in? How loud can it get? Does it duck under the mic? Does it pause when hidden? Does the producer have a mute button? Does the mod queue know when the scene is closed?

If you do not answer those questions, paid moments become random. A viewer pays for a funny TTS, but it fires under a guest answer. A sound alert plays while staff is talking. A browser source reloads and repeats a sound. The fix is not to hate alerts. The fix is to give them a lane.

Build audio modes, not one permanent mix

Audio modes are easier for moderators and producers than constant manual mixing. Set modes for walk, interview, sponsor, food, transit, high-wind, quiet indoor, and fallback. Each mode changes alert rules, music level, TTS availability, and ducking strength.

Walk mode can allow short alerts and TTS if the streamer is mostly chatting. Interview mode should hold long TTS and keep guest audio clear. Food mode may reduce sudden loud sounds because the streamer is close to other people. Transit mode should prioritize route and safety calls. Sponsor mode should block risky browser-source audio entirely.

The viewer-facing menu should match the mode. If paid TTS is closed during an interview, do not let viewers buy it and hope a mod explains later. If sound alerts are held during a restaurant segment, show that they will queue or reopen after the segment.

• Walk mode: short alerts open, music low, mic sidechain active.
• Interview mode: TTS held, music off, guest mic priority.
• Sponsor mode: approved audio only, no surprise browser-source sounds.
• Transit mode: route and safety audio clear, alerts compact.
• Fallback mode: controlled clips or holding sound, live source muted if needed.

Monitor what viewers hear, not what the streamer hears

The streamer is usually hearing the room, not the program. A remote producer should monitor the actual program output and one public playback device when possible. Platform audio processing, destination bitrate, and scene routing can make the viewer mix different from the local monitor.

Twitch's broadcasting guidance and YouTube's encoder guidance both point streamers toward testing and monitoring rather than assuming clean output. For audio, that means doing a real talking test in motion. Read a normal sentence, laugh, let a short alert fire, speak to a guest, and switch scenes. If the producer cannot understand the words, viewers will not either.

Make audio calls short. Mic low. Guest buried. TTS too hot. Wind bad. Music off. Alerts paused. Source clean. Those calls give the streamer information they can act on without turning the broadcast into an audio meeting.

Use ducking gently

Bad ducking is almost as annoying as no ducking. If music slams down every time the streamer breathes, the stream feels broken. If alerts pump under every tiny mic noise, viewers notice the mixer instead of the show. Use attack, release, threshold, and ratio settings that feel natural for speech.

Start with the speech source that matters most and duck only the sources that actually compete. Background music should duck under speech. Long TTS may duck under a guest or pause entirely. A short alert might not need ducking if its volume is already low. Clips should have their own level during fallback so they do not blast viewers after a quiet walking segment.

Keep a before-and-after test clip. Record thirty seconds of walking, guest talking, alert firing, and fallback. If the ducking makes the stream clearer without making the audio jump around, keep it. If it only makes the meters look clever, change it.

Practical StreamableRun setup path

Send the field source into StreamableRun through Moblin, IRL Pro, LiveU, a hardware encoder, or local OBS. Build Cloud OBS scenes first: walk, interview, sponsor, food, transit, fallback, clips, and privacy. Add browser-source alerts only after the base audio is stable.

Then assign audio rules to each scene. Which sources are active? Which sources duck? Which sources are muted? Which sources can a moderator queue? Which source should the producer watch? Keep the public destinations out of the field device so the streamer can focus on the real environment.

Finally, rehearse a messy minute. The streamer walks, a guest speaks, TTS queues, the producer pauses alerts, the source drops, fallback starts, and the source returns. If the audio stays understandable through that, the setup is ready for a real IRL route.

• Source first: get the field mic clean before adding paid audio.
• Scenes second: make different audio rules for walk, interview, sponsor, and fallback.
• Browser sources third: add TTS, alerts, clips, and chat with scene limits.
• Producer controls fourth: mute, pause, replay, and mode switches.
• Destination check last: verify Twitch, Kick, YouTube, or custom RTMP public playback.

Other resources

Use these guides when building the rest of the IRL audio and browser-source workflow around StreamableRun.