Troubleshooting
Symptom first: no captions, missing translation, silent interpreter, unreachable server – the cause and what to do.
This guide starts from the symptom: find in the tables what you see; the right-hand column tells the cause and what to do. The basic setup is in First start.
First the check commands
The --check switch of the caption program only checks, then exits: it writes one block per service the settings require:
livecaption --check
The batch client does the same for the caption server, step by step (FFmpeg, name resolution, TCP port, server, model selection):
captionclient --check --server http://192.168.30.27:5111
The exit code: 0 all fine, 3 FFmpeg, 6 the server is unreachable.
Captions and audio
| Symptom | Cause and what to do |
|---|---|
| No captions although audio plays | With loopback: is the audio really playing on the device you capture? The status bar shows the input level; if it sits around -∞, the device is wrong. |
| Captions stop when nobody speaks | Silent playback is missing: turn it on (Settings → Audio source), or check that the output is not muted. |
| No loopback hears the call program's audio | Switch the audio source to the virtual audio cable; livecaption --check tells whether the cable is present. Details: Teams call with two-way interpreting. |
| With the cable source all sound of the machine is gone | The cable's input became the default playback device (a known side effect of the VB-CABLE installer). Restore the headphones: .\set-default-audio.cmd -Name "…" -Role all. |
| "The server does not support live mode" | An older captionserver: publish it again. If it is new, either --live-max-sessions 0 is set, or the recognition model did not load – the server's start lines tell. |
The connection closes with code 1013 |
The server's live sessions are full. Wait, or start it with a larger --live-max-sessions value. |
| The status bar stays in "Reconnecting…" | The connection broke. The program retries with pauses of 1, 2, 4 … 30 seconds, at most twenty times, and does not restart by itself afterwards. |
Translation and interpreter
| Symptom | Cause and what to do |
|---|---|
| Translation is missing, "–" shows | The translator is unreachable, or the retries ran out. Check with the "Check" button of the settings window. |
| Translation does not start, the band is empty | The speech language and the target language are the same: translation is skipped on purpose. |
| The saved translator key does not work | The settings file comes from another machine or user; the encryption is per machine. Enter it again. |
| The interpreter switch cannot be turned on | Translation is off: without translation there is nothing to say. |
| The interpreter is silent, the dot is amber | The chain is still building: the voice service's model is loading (it can take half a minute), or the in-process voice's model waits for download. Wait it out. |
| The interpreter is silent, the dot is red | The voice service is unreachable, or its model did not load. ..\VoiceServer\start-voiceserver.ps1 -EngineOnly -Check tells which; the front end logs to the voiceserver-log folder, the engine to voice-log. |
| The interpreter captions its own voice | It plays through a loudspeaker. Put on headphones: process exclusion does not protect against sound coming back through the air. |
Services, network and accounts
| Symptom | Cause and what to do |
|---|---|
| The remote client gets a "no answer" error | The Windows firewall drops the incoming connection. At start the server adds the rule by itself, without asking, in the background, for the private and domain networks; this may need administrator confirmation (UAC) – if you do not see its window, check the taskbar, because there is no rule until it is approved. If Windows classifies the network as Public, the default rule does not apply to it: mark the network as private, or – on a trusted network only – delete the rule already added (on the Network page of the web interface) and restart the server with --firewall any – the server does not rewrite an existing rule. Another profile: --firewall private|domain|both|any; to leave it out: --firewall skip. Details: One GPU machine on the network. |
| The client rejects a certificate accepted earlier | The fingerprint changed: the certificate expired, the machine's address changed – or a different machine answers. Compare the fingerprint with the one printed on the server's machine, and accept it again. |
| A service does not start at all: the account service address is missing | The LCP_ACCOUNT_URL variable is mandatory; AccountServer\setup-account-peers.ps1 -Activate sets it. Then restart every open window: the variable does not reach running processes. |
| You cannot sign in to the web admin, but key-based traffic works | The account service or its database is unreachable: while it is down there is no new sign-in, sessions already signed in keep living. Start it, or look at its log; AccountServer\managed-postgres.ps1 -Status shows the state of the database. |
| The account service does not start at all: the database connection string is missing, or "There is no default binding" | The connection string lives in the variable of a binding (LCP_DB_<X>), and the LCP_DEFAULT_DB pointer names the default one (its value is the name of the binding). For an own local database, AccountServer\managed-postgres.cmd creates the database and writes the binding (LCP_DB_LOCAL) and the pointer; for an external PostgreSQL, the db-setup or the db-join maintenance launcher. For an existing binding: AccountServer\maintenance\db-default -Name <NAME>. |
The account service stops because of the old LCP_DB or LCP_KEYRING variable |
Since step 16 there is no base name: run rename-environment-variables.cmd in the repository root (first with -DryRun), then open a new window; reinstall an installed service with an explicit -DbEnv. |
| The account service stops after a minute of waiting: the database is unreachable | The PostgreSQL container is not running, or Docker Desktop is not started. AccountServer\managed-postgres.cmd starts it; the -Status switch tells what the situation is. If Docker Desktop stops with "Virtualization support not detected", WSL is missing on Windows: check-environment.cmd -Fix -Components account. |
| The account service reports a wrong database password | If you just changed it (managed-postgres.ps1 -ResetPassword), open a new window: windows already open keep the old value. Otherwise -ResetPassword gives a new password. |
The account service warns about the old account.db file |
Since 2026-09-23 the database is PostgreSQL, and the old SQLite file is not read: users, keys and peer tokens have to be issued again. The old files can be deleted, or kept for a possible step back. |
| The account service stopped while running: "the database changed" | The database was restored or recreated: start it again. If the services then reject every request, the new database does not know their old peer token – issue a new one: AccountServer\setup-account-peers.ps1 -Replace. |
| The admin password is lost | On the account service's machine: .\start-accountserver.ps1 -UserPassword admin. Details: Accounts, keys and permissions. |
| The voice service launcher reports a missing engine environment | The virtual environment and torch are created by the preparer (VoiceServer\prepare-engine.ps1), because it asks: a download of several gigabytes. The launcher offers it; in a non-interactive run, run it by hand. |
To start the voice service this one command is enough; on the first start it asks:
.\start-voiceserver.ps1
If you cannot find your symptom, write on the Contact page – the service's log and the --check output help a lot.