Best OBS Audio Settings for Recording Voice

People forgive a soft-focus picture. They will not sit through fifteen minutes of thin, boomy, inconsistent audio, and they leave without being able to tell you why.

OBS ships with defaults chosen for a live stream on a modest machine, which is a different problem from recording a voice you intend to keep. Streaming settings optimise for a fixed bandwidth budget and a mix that has to be right the first time. Recording settings can be more generous, and should leave you something to work with afterwards. This page is the second set.

The Short Version

The rest of this page is why each of those, and what breaks when it is wrong.

Settings → Audio

Sample Rate: 48 kHz. Video is a 48 kHz world. Every camera, every editor, every delivery platform expects it. A 44.1 kHz project mixed with 48 kHz sources is a well-known source of slow drift, where audio starts in sync and is visibly late twenty minutes later. If that has already happened to you, OBS audio out of sync has the diagnosis.

Channels: Stereo. This is the channel layout of the whole mix, not of your microphone. Leave it at stereo so system audio, music and any stereo source stay intact. Your microphone gets downmixed to mono separately, further down this page. Setting the global mix to mono to fix a one-sided microphone is a heavy-handed fix that also collapses everything else.

While you are here, choose your devices explicitly rather than relying on Default. OBS captures the device you named, so "Default" following a Bluetooth headset that connected halfway through your session is a genuinely confusing failure to debug.

Settings → Output

Switch Output Mode to Advanced and open the Recording tab. Simple mode hides both of the things that matter here.

Audio Bitrate: 192 kbps. AAC at 160 kbps is fine for speech and is the default. 192 kbps is the better recording default because it leaves headroom for what comes next: every edit, EQ move and re-export is another lossy generation, and bitrate not spent at capture cannot be recovered. 320 kbps is the ceiling and costs roughly 144 MB an hour per track. For a voice you are going to edit, that is cheap.

Audio tracks: 1 and 2. Write your mixed audio to track 1 and your microphone alone to track 2. It costs one checkbox and a little file size, and it is the difference between rebalancing a recording in an editor and re-recording it. The routing is in Advanced Audio Properties and the whole subject is in how to set up multiple audio tracks in OBS.

Recording format. Use Hybrid MP4 on OBS 30.2 and later. It survives a crash the way MKV does and opens everywhere the way MP4 does, and it carries multiple audio tracks. Plain MP4 writes its index last, so a crash usually costs you the entire take.

Mono, and Why Your Voice Is in One Ear

If your voice sits entirely in the left channel, your microphone is arriving as a stereo device with signal on only one side. This is normal for an XLR microphone into an interface, where input 1 becomes the left channel.

Fix it per source, not globally. Right click in the Audio Mixer → Advanced Audio Properties, and tick Downmix to Mono on the microphone row. That folds your voice evenly into both channels and leaves every other source alone.

There is a second reason to do this even when your voice is already centred. A mono voice track is immune to phase problems, plays identically on a phone speaker, and gives an editor one waveform to work with instead of two. There is no benefit to a stereo speaking voice from a single microphone.

Set the Level Before Any Filter

Speak at the volume you will actually present at and watch the meter.

• Peaks at -12 dB to -6 dB, in the yellow band. This is the target.
• Constantly green, nothing reaching yellow means the signal is too quiet, and every later gain move amplifies the room along with your voice.
• Anything touching red is clipping. Clipping is not loudness, it is permanent distortion, and no filter reverses it. This is the one mistake that ends a take.

The fader in the OBS mixer is an output level applied after capture, not gain. If your voice is arriving quiet, raise the gain at the source: the knob on the microphone, the preamp on the interface, or the input level in Windows Sound Control Panel or macOS System Settings → Sound → Input.

The Filter Chain for a Voice

Right click the microphone's gear icon → Filters, then add these in this order. Order changes the result, so it is not a matter of taste.

Noise Suppression first so that everything downstream measures a clean signal. RNNoise is the right default; Speex is lighter on CPU and more aggressive on your voice; the NVIDIA option is excellent if you have the hardware for it.

Gate second, because gating a signal that still contains hum forces the thresholds up. Gate settings depend entirely on your room, and how to add a noise gate to OBS is the measurement procedure. If you speak softly or find any gate audible, use an Expander here instead.

Compressor third. OBS defaults to a 10:1 ratio, which is a limiter wearing a compressor's name. For a voice, 3:1 or 4:1 with the threshold around -20 dB is far more natural. Then use Output Gain to bring the compressed signal back up, typically +3 to +6 dB, so compression makes you consistent rather than quiet.

Limiter last, as a safety net. With the threshold at -1.5 dB nothing you do can produce a clipped export, which matters because you are recording rather than watching a stream go by.

A note on order that catches people out: a compressor placed before a gate lifts the room floor toward your speaking level, which is exactly the distinction the gate is trying to make. Gate then compress, never the reverse.

The Number Almost Nobody Sets

Everything above governs how your voice sounds. Loudness governs whether anyone hears it at the same volume as the video they watched before yours, and OBS has no meter for it at all.

The convention for spoken video is roughly -16 to -14 LUFS integrated, with true peaks below -1 dBTP. YouTube normalises anything louder down toward about -14 LUFS but does not turn quiet uploads up, which is why an under-level tutorial simply sounds quiet next to everything else on the platform.

Because OBS shows peak level rather than loudness, the only way to know is to measure after export. Any editor with a loudness meter will tell you, and so will a free analyser. Measure one representative recording once, note how far off your chain lands, and correct it with the compressor's output gain. After that the chain holds and you never have to think about it again.

The Settings That Are Not Settings

The largest quality gains available to you are not in OBS at all.

• Distance. Speak from 15 to 20 cm away, roughly a spread hand. Further away and the room becomes a larger part of the recording than your voice, which is the single most common reason home audio sounds amateur.
• Angle. Speak slightly across the microphone rather than straight into it. Plosives lose most of their punch when they are not aimed directly at the capsule.
• Direction. A cardioid microphone rejects what is behind it. Point its back at the noisiest thing in the room, which is usually your computer.
• Reflections. A bare desk, a bare wall and a window make a small echo chamber. A rug, a curtain and something soft behind the microphone do more than any filter you can add.

No compressor setting compensates for sitting a metre from the microphone in a hard room.

Streaming Settings Are Not These Settings

If the same OBS profile is also used for live output, three things change.

• Streaming uses one audio track selected in Settings → Output → Streaming, so all your routing work applies to the recording only.
• Streaming bitrate is part of a fixed bandwidth budget, so 128 to 160 kbps is normal there while 192 kbps stays appropriate for the file.
• Live has no second pass, so the gate and compressor have to be conservative. A recording can afford a gentler chain because anything left over can be fixed afterwards.

Keeping two profiles is easier than compromising one. Profile → New in the menu bar.

The Honest Ceiling

Every setting on this page is a one-time job, and once it is right it stays right. It is also a sample rate, a bitrate, a track matrix, a downmix, a gain stage, four filters and a loudness measurement standing between a person and a five minute walkthrough.

Cubix Capture assumes the opposite. No chain to assemble, no routing, no loudness pass: it takes voice and system sound in one go, then works on pacing once there is something to work on, pulling dead air and filler out of a finished take instead of trying to intercept them live. Windows and macOS, with a free tier.

If you are broadcasting live, everything on this page is worth the afternoon, because live output has no second chance.

Keep Reading

• Device, permission and gain, the work that has to happen before any of this.
• The one filter you have to measure rather than copy from a video.
• Suppression methods compared: RNNoise, Speex and the NVIDIA option.
• The track layout referenced above, set up end to end.
• Drift caused by a mismatched sample rate, and how to reverse it.