Audio configuration

Audio is the number one thing people fiddle with in voice chat. This page walks every control under Settings, Voice, what it does, when to change it, and the trade-off.

Tip

Hidden controls? Some advanced sliders only appear in Expert mode. Turn it on under Settings, Advanced, Expert Mode.

Input and output devices

Two side-by-side dropdowns at the top of the panel:

• Input Device is your microphone. System default picks whatever the operating system is currently using. Pick a specific device by name if you have multiple mics.
• Output Device is where audio is played. Same logic.

Each has a volume slider from 0% to 200%. Past 100% the app amplifies digitally, so it can clip if pushed too high.

Caution

Some USB headsets expose two output endpoints (game and chat). Pick the game or handset endpoint for Mumble. The chat endpoint is mono and narrow-band on some products.

Activation mode

Three radio buttons decide when you transmit:

Voice Activation

Mic transmits when speech is detected. Threshold driven, default. The noise removal runs on the input.

Continuous

Always transmit. No noise removal, no gate. Use only in studio rooms or for recording.

Push to Talk

Transmit only while a key is held. Pick the key with the PTT key recorder below.

PTT key recording

When push to talk is selected a shortcut recorder appears. Click it, press the key combination you want, then release. Most modifier combos work (Ctrl+Alt+Z, mouse side buttons, single keys, and more). The key works globally, even when the app is not focused.

Noise removal

Available only in Voice Activation mode. Push-to-talk and Continuous bypass the noise remover. Pick from a dropdown:

Algorithm	Speed	Quality	Best for
DeepFilterNet	Heavy	Best	Modern computers, keyboard or family noise
RNNoise	Light	Very good	Most setups, default
OMLSA-IMCRA	Medium	Good	Smooth output, AC hum
Spectral subtraction	Very light	Basic	Steady backgrounds
None	n/a	n/a	Studio mic in a quiet room

You can switch live. Test by speaking while a steady noise source is running (fan, AC) and compare.

Note

Algorithms that were not compiled into your build will not appear in the dropdown.

Expert: per-algorithm controls

In Expert mode an Advanced controls strip appears for each algorithm. Common knobs:

• Attenuation limit: how much noise the algorithm is allowed to remove.
• Noise floor adapt rate: how fast the algorithm learns new noises.
• Smoothing window: how many frames are averaged for the cleanup.

The defaults are tuned per algorithm. Do not change them unless you can hear the difference and have time to A/B test.

Auto sensitivity

A toggle that continuously re-fits the voice-activation threshold to your room. Recommended for most users.

Calibrate (live level meter)

Hit Calibrate under the threshold slider. Three visual elements appear:

• Fill bar: instant level.
• Peak indicator: a small marker that decays over time.
• Threshold line: a vertical line at the current threshold.

Speak naturally, the fill should comfortably cross the line. Stay silent, the fill should drop below. Adjust the threshold (or turn on auto sensitivity) until both are true.

Audio Processing, Auto Gain

Control	Range	Default	Description
Auto Gain toggle	n/a	On	Pushes quiet speech up to a target loudness.
Max Amplification	1 to 40 dB	20 dB	Ceiling on how much the auto-gain may add.

A higher max can rescue distant speakers, but also makes background noise louder. Pair with a stronger noise removal algorithm.

Compression

Control	Range	Default	What
Quality (bitrate)	8 to 320 kbps	56 kbps	Higher is better quality and more bandwidth.
Audio per packet	10, 20, 40, 60 ms	20 ms	Smaller is lower latency and more packets per second.

Tip

On cellular or shared Wi-Fi, 40 or 60 ms at 24 to 32 kbps is the most resilient. On a wired connection bump to 96 kbps and 20 ms for music-quality voice.

Network

UDP (default)

Voice flows over UDP for minimum latency. This is what you want.

Force TCP

Sends all audio through the existing control connection. Adds a little latency, but passes through any firewall that allows the initial connect. Toggle this under Voice, Network.

Expert section

Visible only when Expert mode is on.

Gate Close Ratio (hysteresis)

The gate has two thresholds, open and close. The close threshold is the open threshold times this ratio (default 0.8). Lower ratio is more aggressive cut-off, and possibly clips the ends of sentences.

Hold Frames

Number of frames to keep the gate open after audio drops below the close threshold. Each frame is the audio-per-packet duration. Increase if your t/s/p/k sounds get cut.

Legacy audio backend Desktop only

The default audio backend supports modern Linux (PipeWire and Pulse), Windows, and macOS. Toggle this to fall back to the legacy backend for niche setups. Takes effect on the next voice toggle.

Audio statistics Expert

Live packet counters per connection:

To Client (server sent):    good   late   lost (n%)   resync
From Client (we received):  good   late   lost (n%)   resync

• good: packets received in order.
• late: out of order, or older than the buffer head.
• lost: never arrived.
• resync: codec resync events after a long pause.

Sustained more than 2% loss is bad. Try Force TCP or a lower bitrate.

Quick recipes

Gaming on a noisy keyboard

Activation: Voice. Noise removal: DeepFilterNet. Bitrate: 64 kbps. Frame: 20 ms.

Podcasting

Activation: Continuous. Noise removal: None. Bitrate: 128 kbps. Frame: 10 ms. Use a hardware pop filter.

Cell-data mobile

Activation: Voice. Noise removal: RNNoise. Bitrate: 24 kbps. Frame: 60 ms. Force TCP on.

Streaming + Discord overlay

Activation: Push to Talk. Noise removal: RNNoise. Bitrate: 56 kbps. Pick a side-button mouse for the PTT key.

Still broken?

Continue to Audio problems.