Voice chat

Players in a network session can talk to each other. It is built in and off by default: tick File > Project Settings > Voice Chat, or set VoiceChat.Enabled = true from a script, and that is all a game has to do. The engine opens the microphone, compresses speech with Opus, sends it through the server and plays it on the other players’ machines. It works over both transports (direct and Steam) and needs nothing installed.

A game with voice chat off never opens a microphone. A game with no network session has nobody to talk to, and does not open one either (except for the microphone test below).

Open microphone or push to talk

The first three are also in Project Settings (Voice Chat, Push To Talk, Push To Talk Key), which is what a new game starts with. All of them can be changed while the game runs, so an options menu can let each player choose:

VoiceChat.PushToTalk = pushToTalkBox;
VoiceChat.PushToTalkKey = chosenKey;        // for example from a "press a key" prompt
VoiceChat.Activation = sensitivitySlider;   // show VoiceChat.MicrophoneLevel beside it

The settings are not saved for you; store them with the rest of the player’s options.

Who hears whom

Team and distance are decided by the server: speech is simply not sent to players who should not hear it, so a modified client cannot listen in on the other team or on a far-away conversation. Proximity and Team Only are in Project Settings too. With Proximity on, voices are ordinary positioned sounds, so they are occluded by walls, pick up reverb zones and use HRTF if it is on (see Audio).

Volume, muting and who is talking

VoiceChat.AutoLevel = true;             // default: even out quiet and loud microphones
VoiceChat.SetPlayerVolume(clientId, 2); // boost one player locally (0 to 4; 1 is normal)
VoiceChat.Volume = 0.8f;                 // other players' voices
VoiceChat.MicrophoneVolume = 1.5f;       // your microphone, before it is sent
VoiceChat.Muted = true;                  // stop sending without turning voice chat off
VoiceChat.Deafened = true;               // stop hearing others
VoiceChat.Mute(clientId, true);          // silence one player, on this machine only
Audio.SetGroupVolume(.Voice, 0.5f);      // the Voice mixer group applies as well

if (VoiceChat.IsSpeaking(clientId)) { /* draw a speaker icon over their head */ }
if (VoiceChat.IsTransmitting) { /* draw a microphone icon for yourself */ }
VoiceChat.OnSpeakingChanged.Add(new (clientId, speaking) => { /* update a player list */ });

Received voices are automatically leveled independently toward a common speech volume, with up to 8× boost and a peak ceiling before the mixer. Silence does not increase the gain. Leveling happens before proximity attenuation, so distance still makes voices quieter. VoiceChat.AutoLevel = false disables leveling; the peak ceiling remains. Individual volumes last for the session; games can expose these controls in their options or player list. Manual volume boosts and the combined mixer can still exceed the peak ceiling. Very faint speech below the open microphone activation threshold needs a lower Activation or higher MicrophoneVolume to be transmitted.

A player’s client id is ownerId on their Network Identity; the host’s own player is 0.

A microphone test for the options menu

VoiceChat.Enabled = true;
VoiceChat.Loopback = true;               // you hear yourself; works without a session
float level = VoiceChat.MicrophoneLevel; // 0 to about 1, for a meter
bool hasOne = VoiceChat.HasMicrophone;   // false: hide the microphone options

Turn Loopback off again when the player leaves the menu.

What it costs

Speech is sent in 20 millisecond frames at 24,000 bits per second: about 3 KB/s for each person talking, and nothing while they are silent. VoiceChat.Bitrate changes it (12,000 is still clear speech; 64,000 is close to transparent). Voice travels on the unreliable channel so that a lost packet is never waited for. Each packet carries a low-quality copy of the one before it, so a single lost packet is rebuilt, and longer gaps are smoothed over.

The server relays speech and never decodes it, so a dedicated server spends almost nothing on voice. It turns away a client that sends far more than a microphone could.

Limits

• No echo cancellation or noise removal. A player on speakers with an open microphone will send the game’s sound back to everyone. Push to talk avoids it; so do headphones.
• Through the server only. There are no private channels, and no talking to players in a lobby before a session exists.
• The default microphone of the system is used; there is no device picker.
• The microphone stays open while voice chat is on and you are in a session, even with push to talk, so that the first word is not clipped. Nothing leaves the machine until you transmit.
• With Steam, this uses Tallow’s own voice path, not Steam’s voice API, so Steam’s voice settings do not apply.

Tests