Wumble

Wumble is a Crystal server that bridges a Mumble connection to a browser through WebRTC using libdatachannel. The browser page is deliberately small, custom JavaScript.

Every Mumble Opus voice packet is put into an RTP packet and sent on the WebRTC audio track assigned to its Mumble session ID. Wumble never decodes, mixes, or combines speaker audio. A browser receives an independent MediaStreamTrack (and audio element) for each speaker. When a Mumble user joins after connection, Wumble renegotiates WebRTC to add that speaker's audio section.

Prerequisites

Install Crystal and the libdatachannel C library and headers through your operating system package manager. The library must be visible to the linker as libdatachannel.

crystal spec
crystal build src/wumble.cr --release
./wumble --bind 0.0.0.0 --port 8080

Open http://gateway-host:8080/ (serve Wumble behind HTTPS in production for iPhone microphone access, audio autoplay, and secure WebSocket access). Connect sends the captured microphone as a WebRTC Opus stream; Wumble forwards its Opus payloads directly to Mumble without decoding or mixing. The connection fields are saved in the URL fragment when their inputs lose focus; fragments are not sent to the server, but passwords remain visible in browser history and copied links. Mumble connection state and control packet names are written to stderr. Set WUMBLE_DEBUG=1 to print backtraces and periodic WebRTC diagnostics, including selected ICE addresses, browser receiver callback/pipe/forwarding counts, accepted Opus sample counts, and track send-buffer sizes.

Network and deployment

The gateway must be able to make TLS TCP connections to the selected Mumble server. libdatachannel also needs UDP reachable between the browser and gateway for WebRTC media. For internet clients, put the gateway on a public address or configure a TURN service in the libdatachannel configuration before deployment.

Mumble voice is carried as encrypted native UDP using the CryptSetup key material. Wumble sends encrypted UDP pings after authentication and accepts voice only after the server replies over UDP. If that path is unavailable after three seconds, the browser is alerted and no TCP UDPTunnel voice is forwarded. Ensure UDP on the selected Mumble port is reachable from the gateway.

Design

• src/wumble/mumble.cr: TLS Mumble control connection and tunnel voice parser.
• src/wumble/datachannel.cr: minimal libdatachannel C binding and one RTP track per speaker.
• src/wumble/server.cr: HTTP static files and JSON WebSocket signalling.
• web/: custom browser UI; it does not import Mumble Web or any other Mumble client.

No Node build, generated web assets, or submodules are used.