Getting clean audio is the single cheapest way to make a stream feel professional. Viewers will forgive a 720p camera long before they forgive clipping, hiss, or a doubled echo. The good news: a basic audio interface plus correct configuration beats most USB “streaming mics,” and most of the work is in software, not gear. This guide covers what to prioritize — correct gain staging first, buffer and driver settings second, processing (noise gate, compressor, EQ) third — and gives exact values for OBS Studio on Windows 11, which is where nearly all streaming audio problems live.
Before touching any slider, do three things: connect the interface via a rear motherboard USB port (front-panel headers and unpowered hubs cause dropouts), install the manufacturer’s ASIO driver if one exists (Focusrite, MOTU, Behringer, Audient all ship them), and set the interface’s sample rate to 48 kHz everywhere — Windows, the driver panel, and OBS. A mismatch is the number-one cause of crackle and pitch artifacts.
Quick Recommended Settings
| Setting | Recommended value | Why |
|---|---|---|
| Sample rate (Windows + driver + OBS) | 48,000 Hz | Streaming platforms deliver 48 kHz; a 44.1/48 mismatch causes resampling artifacts. |
| Bit depth (driver panel) | 24-bit | More headroom below clipping; 16-bit punishes hot gain. |
| Mic gain | Peaks at −18 to −12 dBFS | Leaves headroom for laughter/shouting; never let peaks hit 0 dBFS. |
| OBS sample rate | Settings → Audio → 48 kHz | Must match the OS/device to avoid crackle. |
| Buffer size (ASIO/WASAPI) | 256 samples (128 if CPU allows) | Balances latency against dropouts. |
| Windows sample rate | Sound Control Panel → device → Advanced → 24-bit, 48000 Hz | Single source of truth for the shared-mode clock. |
| Mic monitoring (direct/direct monitoring) | Off for headphones that leak, on for open-back comfort | Hardware zero-latency monitoring; avoid software monitoring delay. |
| Filters in OBS | Noise Suppression (RNNoise) → Noise Gate → EQ → Compressor → Limiter | Order matters; limiter must be last. |
| Compressor ratio | 4:1, threshold −18 dB, output gain to taste | Evens out dynamics without pumping. |
| Limiter threshold | −6 dB | Hard safety ceiling so nothing clips the encoder. |
Setting-by-Setting
Sample Rate: 48 kHz Everywhere
Set it in three places: the interface’s own control panel (e.g., Focusrite Control → Device Settings → Sample Rate → 48 kHz), Windows Sound Control Panel → Recording device → Properties → Advanced → “24-bit, 48000 Hz (Studio Quality),” and OBS → Settings → Audio → Sample Rate → 48 kHz. The trade-off: 48 kHz is what Twitch and YouTube ingest, so anything else forces real-time resampling. Avoid ASIO in OBS unless your interface has no multi-client WASAPI driver — ASIO locks the device from other apps like Discord.
Mic Gain: Peaks at −18 to −12 dBFS
Speak at your normal stream volume while watching OBS’s mic meter. Aim for peaks in the yellow (−18 to −12 dBFS); the meter should never touch the red. The trade-off: conservative gain means quieter output that you recover with compressor makeup gain (cleaner) versus hot gain, which captures more room noise and clips the preamp with no recovery. Most interfaces want the gain knob between 9 and 2 o’clock for dynamic mics like an SM58 or PodMic, and much lower for condensers.
Buffer Size: 256 Samples to Start
In the ASIO control panel (or OBS’s device advanced properties if using WASAPI — Windows exposes buffer indirectly through “Audio Monitor” latency), set 256 samples. Lower to 128 only if you monitor your voice through the PC and hear no crackling. The trade-off: smaller buffers reduce the delay you hear in your headphones, but each drop below your CPU’s capacity produces audible clicks. If you never software-monitor, you can leave buffers generous at 512 and get rock-solid stability.
Noise Suppression: RNNoise at 0 dB
OBS → Filters on your mic source → add Noise Suppression → Method: RNNoise. Suppress the fan/hum first with RNNoise before gating, because a gate on top of steady noise makes gating artifacts obvious. The trade-off: NVIDIA Broadcast-style suppression (if you have an RTX GPU) removes more noise but can eat word endings; RNNoise is gentler and CPU-cheap.
Noise Gate: −40 dB Open, −50 dB Close
Open threshold around −40 dB and close threshold −50 dB works for a treated room; in an untreated room with real noise floor, raise both by 5–10 dB until the gate closes when you stop talking. Attack 5 ms, hold 200 ms, release 150 ms. The trade-off: an aggressive gate chops the tails of words and breath sounds become unnatural — if listeners notice the gate, the thresholds are too high.
Compressor: 4:1, Threshold −18 dB
Ratio 4:1, threshold −18 dB, attack 6 ms, release 60 ms, output gain 3–5 dB. Watch the meter: you want 3–6 dB of gain reduction during normal speech. The trade-off: more compression sounds “radio loud” but pumps on loud laughter; less sounds dynamic but quiet phrases vanish under game audio.
Limiter: −6 dB Threshold, Always Last
Add the Limiter filter after everything else with threshold −6 dB and release 60 ms. This is your insurance policy against a shout clipping the stream. There’s no real downside — it’s transparent until it saves you.
By Hardware Tier
Low Tier: Budget 2-in/2-out Interface (Behringer UMC22, M-Audio M-Track Solo)
These use generic or lightly modified drivers, so skip ASIO in OBS and use WASAPI with the Windows device set to 48 kHz. Keep buffers at default, run RNNoise to compensate for noisier preamps, and push gain carefully — budget preamps get harsh above roughly 80% of travel. Use the hardware direct-monitor switch and keep software monitoring off entirely.
Mid Tier: Focusrite Scarlett Solo/2i2, Audient iD4
Install Focusrite Control 2 or the iD software, set 48 kHz/24-bit in the panel, and enable Direct Monitor for zero-latency self-monitoring. Use WASAPI in OBS so Discord and your browser can share the device simultaneously. You can apply the full OBS filter chain without worrying about preamp noise; the Air mode on Scarlett (if your mic is dull) can substitute for a high-shelf EQ at +2–3 dB.
High Tier: MOTU M2/M4, RME Babyface Pro, Apollo Twin
RME owners get the best driver ecosystem on Windows — use ASIO in OBS with buffers at 128 samples and near-zero dropouts, and use TotalMix FX to apply compression and EQ at the hardware level, offloading the CPU. MOTU M-series owners get excellent loopback: use it to route game audio without a virtual cable. With loopback-capable interfaces, you can eliminate VoiceMeeter entirely, removing one whole layer of routing failure.
Common Mistakes
Crackling and Popping Audio
What causes it: sample-rate mismatch between Windows and the interface, or USB bandwidth/power issues.
How to check: play any audio and listen on headphones; crackle that syncs with CPU spikes points to buffers, constant crackle points to rate mismatch. Verify Windows Sound Control Panel → Properties → Advanced matches the driver panel exactly.
What to do: set everything to 48 kHz, move the cable to a rear USB port, and disconnect other high-bandwidth USB devices sharing the same controller.
How to undo: revert each sample-rate location to its previous value individually so you can isolate which mismatch mattered.
Doubled/Echoed Voice for Viewers
What causes it: the same microphone feeding the stream twice — usually a “Listen to this device” route plus OBS, or monitoring routed back into a capture.
How to check: in OBS, mute the mic source; if viewers still hear you, a second path exists. Check Sound Control Panel → Recording device → Properties → Listen tab.
What to do: uncheck “Listen to this device,” and confirm Discord/voice apps use the interface for input but don’t feed output into a desktop-audio capture.
How to undo: re-enable Listen only if you genuinely need local monitoring without OBS running.
Voice Too Quiet Compared to Game Audio
What causes it: game/desktop audio at default 100% while mic relies on preamp gain alone, and OBS’s Advanced audio properties left at defaults.
How to check: in OBS → Edit → Advanced Audio Properties, compare desktop audio (typically −4 to −8 dB) against mic levels in the mixer.
What to do: set mic at roughly −2 to 0 dB and desktop audio at −8 to −12 dB, with compressor makeup gain doing the heavy lifting on the mic. Never compensate by pushing preamp gain into clipping.
How to undo: restore both to 0 dB and rebalance from the source apps’ volumes instead.
Chipmunk or Slowed Voice Recording on VODs
What causes it: OBS recording at 44.1 kHz while devices run 48 kHz, or a device added to OBS that later disconnected, forcing resampling of silence.
How to check: compare Settings → Audio sample rate against every device’s Properties → Advanced rate.
What to do: set OBS and all devices to 48 kHz; remove unused audio devices from OBS’s audio mixer entirely.
How to undo: recordings already made can be slowed/corrected in an editor by resampling, but it’s lossy — fix the source going forward.
FAQ
Do I need an audio interface at all, or is a USB mic fine?
A USB mic is fine to start. An interface wins when you want an XLR dynamic mic (far better rejection of keyboard and room noise), hardware zero-latency monitoring, and a physical gain knob you can ride mid-stream. If your current USB mic peaks cleanly at −18 dBFS with RNNoise applied, upgrading adds less than fixing your room does.
Should I use ASIO in OBS?
Only if your interface’s ASIO driver is excellent (RME, MOTU) and you don’t need Discord or a browser to share the mic. ASIO is exclusive-mode: OBS will block other apps from the device. For Focusrite Scarlett and most budget units, the multi-client WASAPI path is more practical and the latency difference is inaudible for streaming.
What sample rate and bitrate should I stream at?
48 kHz, always. Twitch’s audio codec operates at 48 kHz, and OBS audio bitrate at 160–320 kbps is transparent for voice-plus-music. Audio quality problems almost never come from bitrate — they come from gain staging and rate mismatches.
How do I hear myself without delay?
Use your interface’s hardware direct monitoring — the “Direct Monitor” switch on a Scarlett, or the monitor mix knob on a MOTU — so your voice routes to headphones before it ever touches the PC. Software monitoring through OBS (Monitor via Edit → Advanced Audio Properties) adds a buffer-sized delay that most people find disorienting. If you must software-monitor, drop buffers to 128 samples first.
