Speech-dispatcher handling improved.

docs/user.md

@@ -369,6 +369,25 @@ See the [keyboard guide](../config/keyboard/Readme.md) for an example.
 The default driver is `speech-dispatcher`. Select its module, language, and
 voice in Speech settings; empty values use server defaults.
 
+Fenrir uses `SPEECHD_ADDRESS` when set: `unix_socket:/absolute/path` selects a
+Unix socket, and `inet_socket:HOST:PORT` selects TCP (default host `127.0.0.1`,
+port `6560`). Invalid addresses report an error. Otherwise, `SPEECHD_SOCKET`
+selects a Unix socket. Without either variable, Fenrir uses
+`$XDG_RUNTIME_DIR/speech-dispatcher/speechd.sock`, falling back to
+`$XDG_CACHE_HOME/speech-dispatcher/speechd.sock` or
+`~/.cache/speech-dispatcher/speechd.sock` when the runtime directory is unset.
+With sudo, the default remains `/run/user/$SUDO_UID/speech-dispatcher/speechd.sock`.
+Sudo may remove address variables; preserve them explicitly if needed.
+
+When opening speech, Fenrir first tries the selected server. If its socket is
+missing or refuses connections, Fenrir requests `speech-dispatcher --spawn`
+and briefly retries. The daemon's configuration controls whether spawning is
+allowed. `SPEECHD_CMD` can select another executable. Autostart is limited to
+Unix sockets and local loopback TCP servers. When sudo changes the account,
+start Speech Dispatcher as the invoking user first; Fenrir will not launch a
+root-owned daemon for that user. Settings discovery never starts a daemon or
+interrupts the active speaking client.
+
 Hardware driver IDs are `dectalk`, `dectalk-usb`, `litetalk`, `doubletalk`, and
 `tripletalk`. Serial drivers require `presentation.speech.hardware_device`,
 for example `/dev/ttyUSB0`, and default to 9600 baud when