ts3java/ts3-client

TS3J Client

A desktop TeamSpeak 3 client built on top of the ts3j reverse-engineered TS3 protocol library. It looks and behaves like the official TS3 client and focuses first on the features that matter most: real-time voice with Opus encoding and voice-activation detection (VAD).

The project is split so the frontend can be replaced (e.g. a future web UI via TeaVM/CheerpJ) without touching the library:

Module	Artifact	Responsibility
core	ts3-client-core	Frontend-agnostic library: protocol integration, server model, connection orchestration, the voice pipelines and their DSP. No UI, no platform audio API, no native code.
desktop	ts3-client-desktop	Desktop audio backend: PipeWire (or Java Sound) capture/playback + native Opus via the FFM API (project Panama), plus the global hotkey hooks.
swing	ts3-client-swing	Swing desktop UI + entry point. Depends on core and desktop.

The core exposes AudioBackend / VoiceInput / VoiceOutput; the frontend injects a concrete backend (DesktopAudioBackend) into TeamspeakConnection. A different frontend supplies its own UI and audio backend while reusing core unchanged, as the Android app in ../android does.

Features

Voice (the priority)

• Native Opus codec via a direct Panama (Foreign Function & Memory API) binding — no JNA, no other third-party FFI. The system libopus is used when installed, otherwise a bundled copy is extracted from the JAR. Only the Windows x86-64 build is bundled by default (build with -Dnatives.all to package every platform); elsewhere install libopus from your package manager. Encoding at 48 kHz, 20 ms frames.
• Voice Activation Detection (VAD) with the same three modes as the TS3 client:
  • Volume Gate — RMS/dBFS threshold with a live input meter and hangover.
  • Automatic — a Java port of WebRTC's RNN speech detector (rnn_vad), the one the TS3 client uses, with the same weights.
  • Hybrid — transmit only when loud enough and detected as speech.
  • Optional VAD over Push-To-Talk (keep detecting voice while PTT is available).
• Capture pre-processing — a Java port of the WebRTC audio-processing stages the TS3 client enables, in its order: high-pass filter → noise suppression (WebRTC NS, TS3's four levels) → typing attenuation (the transient suppressor, told about key presses) → automatic gain control (AGC2, WebRTC's successor to the AGC1 TS3 uses). Checked against WebRTC built from source.
  • Boost quiet speech (0–20 dB) raises AGC2's cap on the amplified noise floor, which otherwise holds back a distant voice that is barely above the background noise.
  • Echo cancellation (WebRTC AEC3) is intentionally omitted — it needs the loudspeaker reference signal and matters mainly for open speakers, not the typical headset.
• Push-to-Talk — bind any key or mouse button as a global hotkey; transmits only while held.
• Continuous transmission mode.
• Adaptive jitter buffer — a port of the libspeex jitter buffer the TS3 client uses, set up and driven as it does: the delay follows the measured network jitter, and what it has learned carries over between talk bursts.
• Playback mixing — each speaker gets their own Opus decoder and audio line (a stream of their own in the PipeWire mixer), so one slow decode never blocks the others.
• Whisper receive — targeted voice is decoded and played like normal voice.
• Per-client mute, master deafen, adjustable mic gain / playback volume.
• On-the-fly Opus tuning — bitrate, complexity, VBR, FEC and voice/music codec can all be changed live from the options dialog and take effect on the running encoder immediately (no reconnect); a voice↔music switch transparently rebuilds the encoder because Opus fixes its application mode at creation.
• Configurable capture/playback devices.

Server interaction

• Connect to any TS3 server (an identity is generated on first use if none exists).
• Channel/client tree styled like TS3, updated live from protocol events (joins, leaves, moves, channel create/edit/delete, nickname/mute/away changes).
• Talk indicators — speakers turn green live as they talk.
• Info panel — selecting a channel shows its topic and description (fetched on demand); selecting a client shows its server groups and channel group (names resolved from the server's group lists), talk power, platform and version.
• Server-group badge next to each client's nickname in the tree.
• Spacer channels — TS3 [spacer]/[*spacer]/[c/l/r spacer] names render as non-interactive separators (fill, centred, aligned).
• Double-click a channel to join; right-click a client to poke, open a private chat, or locally mute them.
• Chat to the current channel or the whole server; receive channel/server/private messages and pokes.
• Avatars — a selected client's avatar is shown in the info panel, fetched from the server's file repository (/avatar_<id> in channel 0, verified against the client's client_flag_avatar MD5) and cached on disk; animated GIFs play. Self → Set avatar… uploads your own picture (PNG/JPEG/GIF/BMP, within the server's i_client_max_avatar_filesize) or removes it.
• Server bookmarks — quick-connect menu with add/edit/remove management, each optionally pinned to a specific identity.
• Identity management (Tools → Identities) — keep several identities, mark one as the default, pick one per server or per bookmark, rename, raise an identity's security level, and import/export TeamSpeak-compatible .ini identity files.
• Import from TeamSpeak 3 / myTeamSpeak — the official client keeps its bookmarks and identities (the data a myTeamSpeak account synchronises) as protobuf Item_Data blobs in the ProtobufItems table of its settings.db. On first run, when this client has no identity of its own yet, it adopts all of them — identities, the default identity and its nickname, and the bookmarks with the identities they connect with; "Import from TeamSpeak 3…" in Manage Bookmarks and in Identities re-imports at any time (additive: known identities and bookmarks are left alone). The schema was recovered from the descriptors embedded in the client binary.
• Contacts (Tools → Contacts) — TeamSpeak's friend/blocked list: mark a client as Friend, Blocked or Neutral from its context menu (or drag it into the window), give it a custom and a phonetic nickname, choose how the name is shown (custom name and nickname / only custom / only nickname), and per contact mute automatically, ignore server/channel chat, private chat, pokes and away messages, and allow or deny whispers. Friends are coloured green and blocked clients red in the tree (optional), per-type defaults for new contacts are configurable ("Set Defaults"), and the global whisper policy lives in Options → Contacts. The list is stored in ~/.ts3jclient/contacts.txt in the exact record format the official client keeps in the Contacts table of its settings.db; on first run it is seeded from an installed TeamSpeak 3 client's database (read with a small built-in SQLite reader — no driver needed), and can be re-imported any time.
• Per-client volume ("Change Volume…") — a dB gain on top of the master volume, remembered in the client's contact entry.
• Ban List (Tools → Ban List) — the server's bans in a searchable table (Add / Remove / Edit, own-ban filter and highlighting, choosable columns), with the official Add dialog: IP and nickname patterns read as IPv4/IPv6 wildcards, fixed strings or regular expressions, unique id, myTeamSpeak id, reason presets and a duration capped by i_client_ban_max_bantime.
• Self status — Away (with message) and Channel Commander toggles.
• Status bar shows the server name, user count and live ping.
• Change your nickname, mute/deafen from the toolbar.

Hotkeys

• Global hotkeys, as in TeamSpeak 3: single keys or combinations (any key can act as the modifier) and mouse buttons including Mouse 4 / Mouse 5, working system-wide rather than only while the window has focus. Configure them in Options → Hotkeys.
• Per binding you choose whether it triggers on key down or on key up, and whether it applies on the active server only or to every connected one (checkbox off). Push-to-talk style actions ignore the trigger setting and simply last while held.
• The action list is TeamSpeak's own, reverse-engineered from its hotkey dialog: three categories (Server, Self, Misc) and a "Show advanced actions" checkbox that reveals the rest, exactly as the original hides all but the everyday actions. Actions this client does not implement are listed but greyed out.
• Push-to-talk is one of those hotkeys; the button in Options → Voice Activation edits that binding.
• On Linux, capture uses XInput2 raw events, as TeamSpeak's own client does: no privileges needed and the keystroke is not swallowed. X11's RECORD extension is kept as a fallback for servers without XInput2.
• On Windows, capture uses raw input (RIDEV_INPUTSINK) through a message-only window: it observes rather than intercepts, so unlike a low-level keyboard hook it cannot swallow a keystroke or stall the system input queue, and it reports the side buttons and both edges of every key. Bindings are stored as scan codes, so they follow the physical key rather than the layout.
• On Wayland the hooks run inside Xwayland, so they see every X11 application, and keys aimed at native Wayland windows only where the compositor forwards them — under KWin that is System Settings → Window Management → Legacy X11 App Support, which is why TeamSpeak's hotkeys work there. Where it forwards nothing, the client reads /dev/input/event* instead, for which your user has to be in the input group. The Hotkeys tab says which backend is in use, or why none is.

Notification sounds

• Sound packs in TeamSpeak's own format: a folder of waves plus a settings.ini mapping actions (CONNECTION_CONNECTED, CLIENT_MOVED_TO_CURRENT_CHANNEL_STAYS, …) to play("file.wav") entries, with ${clientType}/${channelname}/… placeholders resolved per event. Packs are picked up from ~/.ts3jclient/sound, an installed TeamSpeak 3 client ($TS3_CLIENT_DIR or the usual install paths) and a folder of your choosing, so the official packs work unchanged.
• Per-action configuration (Options → Notifications): switch any action's sound off, and mark actions as important (shown in bold, as in TS3) — important actions are the only ones still played while the speakers are muted.
• Sounds are mixed onto one playback line, so overlapping events never fight over the device; pack volume is separate from the voice volume.

Icon packs

• Icon packs in TeamSpeak's own format: a .zip (or unpacked folder) of SVG/PNG artwork plus a settings.ini mapping icon keys (CHANNEL_GREEN, PLAYER_ON, CONNECT, …) to files. Packs are picked up from ~/.ts3jclient/gfx, an installed TeamSpeak 3 client ($TS3_CLIENT_DIR or the usual install paths) and a folder of your choosing, so the official packs (default_colored_2014.zip, default_mono_2014.zip, the legacy default.zip) work unchanged.
• Material, a pack in the same format built from Google's Material icons, ships with the client (core/src/main/resources/gfx/material, generated by tools/material-icon-pack.py). Its monochrome icons paint with currentColor, which follows the theme's text colour; it covers every icon TeamSpeak's own packs do. It is the Android app's default, and the Android app draws TeamSpeak packs too — import one (e.g. default_colored_2014.zip) in its settings.
• The pack draws the tree (server, channel, client state), the toolbar, the menus, the context menus, the file browser and TeamSpeak's default group icons. Vector art is rasterised at the size it is drawn at; a pack's FALLBACK option decides whether icons it lacks come from default.zip, and anything still missing falls back to the icons the client draws itself.
• Pick one in Options → Design, which also shows a viewer of everything the pack contains; switching redraws the window right away.

Display scaling (Linux)

• Java2D on X11 (and so under XWayland) only scales by whole numbers, so 125 % or 150 % comes out at 100 %. On Linux the client pins Java2D to 1 and hands the full factor to FlatLaf instead, scaling its own sizes and icons to match.
• The factor follows the desktop: XSETTINGS Xft/DPI (GNOME, Plasma), else KWin's XWayland scale, else GDK_SCALE. For crisp text on Wayland, let X11 apps scale themselves (Plasma: Display Configuration → Legacy Applications → Apply scaling themselves; GNOME: the xwayland-native-scaling mutter feature), otherwise the compositor upscales the window.
• Options → Design → Interface scale overrides it (after a restart); -Dflatlaf.uiScale= or -Dsun.java2d.uiScale= on the command line bypasses all this.

Requirements

• Java 25+ (the LTS; developed on Temurin 26)
• The native Opus library on the system:
  • Debian/Ubuntu: sudo apt install libopus0
  • Arch: sudo pacman -S opus
  • macOS: brew install opus

Building

The build compiles ts3j from ../ts3j alongside the client, so make sure the submodule is checked out (git submodule update --init):

mvn -DskipTests package

This produces a runnable fat-jar at swing/target/ts3-client.jar.

Running

java -jar swing/target/ts3-client.jar
# or, during development:
mvn -pl swing exec:java

Then use Connections → Connect…, enter a server address, port (default 9987), nickname and identity, and connect. Open Tools → Options to pick audio devices and tune voice activation while watching the live meter.

Architecture

core/   com.ts3client
├── config                   the profile (~/.ts3jclient): Settings, Bookmarks, IdentityStore,
│                            AwayMessages, …; AppDirs finds it, ProfileFiles writes it
│                            privately and all at once
├── session
│   ├── ServerSession        one server tab's state and actions, whatever the frontend
│   └── Sessions             what the servers share: the microphone, away state, nickname
├── net
│   ├── TeamspeakConnection  ties socket + audio + model; runs requests on its own pool
│   ├── ConnectionEventHandler  server events → model changes, sounds and log lines
│   ├── ServerModel          channel/client state; ChannelNode/ClientEntry view models
│   ├── ChannelAdmin/BanAdmin/AvatarAdmin  channel editing, the ban list, our own avatar
│   ├── IconRepository       server icons (avatar/ holds the client avatar cache)
│   ├── filetransfer         channel file browsing, up- and downloads
│   └── ConnectionListener   frontend callbacks
├── audio                    the voice pipelines every platform shares
│   ├── AudioBackend/AudioIo what a platform supplies: devices, lines, Opus
│   ├── CaptureVoiceInput    capture → processing → VAD/PTT gate → Opus encode
│   ├── StreamingVoiceOutput per-speaker decode, each through a VoiceStream
│   ├── JitterBuffer         port of the libspeex jitter buffer TS3 uses
│   ├── PerSpeakerPlayout    a line per speaker (desktop); MixedPlayout mixes onto one (Android)
│   ├── processing           port of TS3's WebRTC capture chain: HPF, NS, transients, AGC2
│   ├── vad                  port of WebRTC's rnn_vad speech detector
│   └── opus                 codec interfaces
├── chatlog                  TS3-compatible chat logs, shared with the TS3 client by default
├── contacts                 friends and blocked clients, in TS3's own record format
├── text                     BBCode, TeamSpeak links, chat search
├── gfx
│   ├── IconPack             an icon pack zip/folder/classpath folder + its settings.ini mapping
│   ├── IconPacks            discovery of installed packs, the shipped Material one, lookup order
│   └── IconKeys             the icon keys the channel tree asks for
├── hotkey
│   ├── HotkeyAction         catalogue of TeamSpeak's hotkey actions (RE'd from its binary)
│   ├── Hotkey/HotkeyCombo   one binding: keys, trigger edge, server scope, argument
│   ├── Hotkeys              persisted bindings (~/.ts3jclient/hotkeys.properties)
│   ├── GlobalInputHook      platform hook interface: system-wide key/button events
│   └── HotkeyEngine         held-key tracking, combination matching, recording
├── sound
│   ├── SoundEvent           catalogue of actions (TeamSpeak's own event ids)
│   ├── SoundPack            a pack folder + its settings.ini mapping
│   ├── SoundPacks           discovery of installed packs
│   ├── NotificationSettings per-action enabled / important flags
│   ├── SoundNotifier        decides what is heard (pack + config + mute state)
│   ├── SoundPlayer          renders a sound
│   └── PackSoundPlayer      decodes, resamples and mixes sounds onto one line
└── teamspeak                the official client's own data files
    ├── TeamSpeakSettingsDb  locates settings.db; reads Contacts and the ProtobufItems sync store
    ├── SqliteReader         minimal read-only SQLite 3 file reader (no driver)
    ├── SyncItem             a synced bookmark / identity / folder (Item_Data)
    ├── SyncItemDecoder      protobuf wire decoding of Item_Data, via ProtobufReader
    └── TeamSpeakImporter    merges synced items into IdentityStore / Bookmarks

desktop/  com.ts3client.audio.desktop + com.ts3client.hotkey.desktop
├── DesktopAudioBackend      PipeWire where it runs, Java Sound elsewhere, and native Opus
├── Opus/NativeOpusCodec     Panama (FFM) binding to native libopus
├── AudioDevices             device enumeration + line opening (48 kHz/16-bit)
├── JavaSoundCapture/JavaSoundPlayback  Java Sound lines
├── pipewire                 PipeWire streams through FFM
└── hotkey.desktop
    ├── XInput2InputHook     global key/button capture via XInput2 raw events
    ├── XRecordInputHook     the same via X11's RECORD extension, as a fallback
    ├── EvdevInputHook       /dev/input fallback for Wayland sessions
    ├── WindowsInputHook     the same on Windows: raw input into a message-only window
    ├── RawInput             RAWINPUT layout and decoding, split out to be testable
    ├── X11KeyNamer/WindowsKeyNamer  layout-aware key labels per platform
    └── DesktopInputHooks    picks the backend that suits the session

swing/  com.ts3client
├── Main                     entry point (look & feel, settings, backend injection)
└── ui
    ├── MainFrame            window: menu, toolbar, tree | chat, status bar
    ├── ServerTreePanel      TS3-style channel/client tree (group badges, spacers)
    ├── Spacers              TS3 spacer-channel name parsing/rendering
    ├── InfoPanel            channel description / client group + details view
    ├── ChatPanel            chat log + input
    ├── SettingsDialog       the Options dialog; one panel per tab (DevicesPanel, …, ChatLogsPanel)
    ├── HotkeysPanel         hotkey list (Options → Hotkeys)
    ├── HotkeyDialog         add/edit one hotkey: action, combination, trigger, scope
    ├── HotkeyService        bindings + engine + input hook, for the dialogs
    ├── HotkeyActions        carries a fired hotkey out on the client
    ├── NotificationsPanel   sound pack + per-action sound/important configuration
    ├── IconPackPanel        icon pack chooser + icon viewer (Options → Design)
    ├── DisplayScale         interface scale on Linux (fractional, via FlatLaf)
    ├── ConnectDialog        connect form
    ├── BookmarksDialog      manage saved servers
    ├── IdentitiesDialog     manage identities (new/import/export/improve)
    ├── IdentityChooser      identity drop-down shared by connect/bookmark forms
    ├── LevelMeter           dBFS meter with threshold marker
    ├── IconTheme            active icon pack: rasterising (JSVG) and caching
    ├── Icons                icon lookup by TeamSpeak key, with drawn fallbacks
    └── Theme                palette + fonts

Known limitations / next steps

• Whisper is received/played but not yet sendable from the UI.
• Hotkey actions TeamSpeak offers but this client does not perform yet (listed but greyed out in the dialog): capture/playback/hotkey profiles and sound packs, whisper and push-to-whisper, recording, plugins, server groups, talk power, 3D sound, hardware ("local") microphone mute and the channel-traversal variants beyond "Switch to Channel".
• Global hotkeys cover Linux and Windows; macOS has no backend, so hotkeys there are inert (a CGEventTap, which needs Accessibility permission, is the way in). The Windows backend is written but has not been run on Windows — only its decoding is covered by tests.
• When no backend can start, push-to-talk does not work at all, not even with the window focused; a focus-scoped fallback hook would restore the pre-hotkey behaviour.